Skip to contentSkip to conversation

Documentation: remove the requirement for LLM attribution

PATCH In review Jeff Layton 16 participants 69 messages Jul 2 – Jul 22

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Documentation: remove the requirement for LLM attribution

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

2/ It basically serves as free advertising for proprietary LLM companies.

3/ It's not clear why we want to collect this info in the first place.

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

---
Christian had proposed watering down the LLM attribution, but I think it's not productive to try and track this until we have a clearer sense of what we want to do with this information and how to make it more reliable.

---
 Documentation/process/coding-assistants.rst  | 22 ----------------------
 Documentation/process/submitting-patches.rst | 10 ----------
 2 files changed, 32 deletions(-)
Documentation/process/coding-assistants.rst +0 -22
index 899f4459c52d..c4cc0917fc92 100644
@@ -35,25 +35,3 @@ is responsible for:
 * Ensuring compliance with licensing requirements
 * Adding their own Signed-off-by tag to certify the DCO
 * Taking full responsibility for the contribution
-
-Attribution
-===========
-
-When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
-
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
-
-Where:
-
-* ``AGENT_NAME`` is the name of the AI tool or framework
-* ``MODEL_VERSION`` is the specific model version used
-* ``[TOOL1] [TOOL2]`` are optional specialized analysis tools used
-  (e.g., coccinelle, sparse, smatch, clang-tidy)
-
-Basic development tools (git, gcc, make, editors) should not be listed.
-
-Example::
-
-  Assisted-by: Claude:claude-3-opus coccinelle sparse
Documentation/process/submitting-patches.rst +0 -10
index cc6a1f73d7f2..b74c38aa9770 100644
@@ -634,16 +634,6 @@ bugzilla.kernel.org is a public place in this sense, but email addresses
 used there are private; so do not expose them in tags, unless the person
 used them in earlier contributions.
 
-Using Assisted-by:
-------------------
-
-If you used any sort of advanced coding tool in the creation of your patch,
-you need to acknowledge that use by adding an Assisted-by tag.  Failure to
-do so may impede the acceptance of your work.  Please see
-Documentation/process/coding-assistants.rst for details regarding the
-acknowledgment of coding assistants.
-
-
 .. _the_canonical_patch_format:
 
 The canonical patch format

---
base-commit: 665159e246749578d4e4bfe106ee3b74edcdab18
change-id: 20260702-aidoc-7e18f221d63a

Best regards,

Signature · 1 line
Jeff Layton <[email protected]>
#1 lore raw
Quoted 16 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

2/ It basically serves as free advertising for proprietary LLM companies.

3/ It's not clear why we want to collect this info in the first place.

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

Quoted 8 lines

---
Christian had proposed watering down the LLM attribution, but I think it's not productive to try and track this until we have a clearer sense of what we want to do with this information and how to make it more reliable.

---
 Documentation/process/coding-assistants.rst  | 22 ----------------------
 Documentation/process/submitting-patches.rst | 10 ----------

You're missing scripts/checkpatch.pl.

Quoted 22 lines
 2 files changed, 32 deletions(-)
Documentation/process/coding-assistants.rst +0 -12
index 899f4459c52d..c4cc0917fc92 100644
@@ -35,25 +35,3 @@ is responsible for:
 * Ensuring compliance with licensing requirements
 * Adding their own Signed-off-by tag to certify the DCO
 * Taking full responsibility for the contribution
-
-Attribution
-===========
-
-When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
-
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
-
-Where:
-

Documentation/process/coding-assistants.rst is responsible for: · line 59

Quoted 10 lines
Documentation/process/coding-assistants.rst +0 -10
-* ``AGENT_NAME`` is the name of the AI tool or framework
-* ``MODEL_VERSION`` is the specific model version used
-* ``[TOOL1] [TOOL2]`` are optional specialized analysis tools used
-  (e.g., coccinelle, sparse, smatch, clang-tidy)
-
-Basic development tools (git, gcc, make, editors) should not be listed.
-
-Example::
-
-  Assisted-by: Claude:claude-3-opus coccinelle sparse

Documentation/process/generated-content.rst still clearly requests contributors to disclose usage of code-generation tools. Even if we drop the Assisted-by tag, I think it is useful to mention the requirement here (probably including a link to generated-content.rst).

Quoted 25 lines
Documentation/process/submitting-patches.rst +0 -11
index cc6a1f73d7f2..b74c38aa9770 100644
@@ -634,16 +634,6 @@ bugzilla.kernel.org is a public place in this sense, but email addresses
 used there are private; so do not expose them in tags, unless the person
 used them in earlier contributions.
 
-Using Assisted-by:
-------------------
-
-If you used any sort of advanced coding tool in the creation of your patch,
-you need to acknowledge that use by adding an Assisted-by tag.  Failure to
-do so may impede the acceptance of your work.  Please see
-Documentation/process/coding-assistants.rst for details regarding the
-acknowledgment of coding assistants.
-
-
 .. _the_canonical_patch_format:
 
 The canonical patch format
 
---

base-commit: 665159e246749578d4e4bfe106ee3b74edcdab18
change-id: 20260702-aidoc-7e18f221d63a

Signature · 3 lines
Regards,

Laurent Pinchart
#2 lore raw
Quoted 57 lines · Laurent Pinchart
Quoted 16 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

2/ It basically serves as free advertising for proprietary LLM companies.

3/ It's not clear why we want to collect this info in the first place.

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

Quoted 8 lines

---
Christian had proposed watering down the LLM attribution, but I think it's not productive to try and track this until we have a clearer sense of what we want to do with this information and how to make it more reliable.

---
 Documentation/process/coding-assistants.rst  | 22 ----------------------
 Documentation/process/submitting-patches.rst | 10 ----------

You're missing scripts/checkpatch.pl.

Quoted 29 lines
 2 files changed, 32 deletions(-)
Documentation/process/coding-assistants.rst +0 -19
index 899f4459c52d..c4cc0917fc92 100644
@@ -35,25 +35,3 @@ is responsible for:
 * Ensuring compliance with licensing requirements
 * Adding their own Signed-off-by tag to certify the DCO
 * Taking full responsibility for the contribution
-
-Attribution
-===========
-
-When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
-
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
-
-Where:
-
-* ``AGENT_NAME`` is the name of the AI tool or framework
-* ``MODEL_VERSION`` is the specific model version used
-* ``[TOOL1] [TOOL2]`` are optional specialized analysis tools used
-  (e.g., coccinelle, sparse, smatch, clang-tidy)
-
-Basic development tools (git, gcc, make, editors) should not be listed.
-
Quoted 8 lines

Documentation/process/coding-assistants.rst is responsible for: · line 59

Quoted 3 lines
Documentation/process/coding-assistants.rst +0 -3
-Example::
-
-  Assisted-by: Claude:claude-3-opus coccinelle sparse

Documentation/process/generated-content.rst still clearly requests contributors to disclose usage of code-generation tools. Even if we drop the Assisted-by tag, I think it is useful to mention the requirement here (probably including a link to generated-content.rst).

I mean this is what I proposed, but Christian was opposed to me doing so?

I guess if this patch is going in over my objections anyway you may as well do what I suggested with it, but I was going to send a separate patch providing a link.

Let me know if it's not worth my time doing that.

Quoted 23 lines
Quoted 23 lines
Documentation/process/submitting-patches.rst +0 -10
index cc6a1f73d7f2..b74c38aa9770 100644
@@ -634,16 +634,6 @@ bugzilla.kernel.org is a public place in this sense, but email addresses
 used there are private; so do not expose them in tags, unless the person
 used them in earlier contributions.
 
-Using Assisted-by:
-------------------
-
-If you used any sort of advanced coding tool in the creation of your patch,
-you need to acknowledge that use by adding an Assisted-by tag.  Failure to
-do so may impede the acceptance of your work.  Please see
-Documentation/process/coding-assistants.rst for details regarding the
-acknowledgment of coding assistants.
-
-
 .. _the_canonical_patch_format:
 
 The canonical patch format
Quoted 8 lines
Quoted 3 lines

---
base-commit: 665159e246749578d4e4bfe106ee3b74edcdab18
change-id: 20260702-aidoc-7e18f221d63a

--
Regards,

Laurent Pinchart

Thanks, Lorenzo

#3 lore raw
In reply to Laurent Pinchart
Quoted 57 lines · Laurent Pinchart
Quoted 16 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

2/ It basically serves as free advertising for proprietary LLM companies.

3/ It's not clear why we want to collect this info in the first place.

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

Quoted 8 lines

---
Christian had proposed watering down the LLM attribution, but I think it's not productive to try and track this until we have a clearer sense of what we want to do with this information and how to make it more reliable.

---
 Documentation/process/coding-assistants.rst  | 22 ----------------------
 Documentation/process/submitting-patches.rst | 10 ----------

You're missing scripts/checkpatch.pl.

Quoted 29 lines
 2 files changed, 32 deletions(-)
Documentation/process/coding-assistants.rst +0 -19
index 899f4459c52d..c4cc0917fc92 100644
@@ -35,25 +35,3 @@ is responsible for:
 * Ensuring compliance with licensing requirements
 * Adding their own Signed-off-by tag to certify the DCO
 * Taking full responsibility for the contribution
-
-Attribution
-===========
-
-When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
-
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
-
-Where:
-
-* ``AGENT_NAME`` is the name of the AI tool or framework
-* ``MODEL_VERSION`` is the specific model version used
-* ``[TOOL1] [TOOL2]`` are optional specialized analysis tools used
-  (e.g., coccinelle, sparse, smatch, clang-tidy)
-
-Basic development tools (git, gcc, make, editors) should not be listed.
-
Quoted 9 lines

Documentation/process/coding-assistants.rst is responsible for: · line 59

Quoted 3 lines
Documentation/process/coding-assistants.rst +0 -3
-Example::
-
-  Assisted-by: Claude:claude-3-opus coccinelle sparse

Documentation/process/generated-content.rst still clearly requests contributors to disclose usage of code-generation tools. Even if we drop the Assisted-by tag, I think it is useful to mention the requirement here (probably including a link to generated-content.rst).

Fair point. Maybe we should keep the Attribution section then, and give it some brief text like:

"Contributions developed with LLM help should also follow the guidelines outlined in the generated-content document."

With a link to generated-content.rst?

Quoted 25 lines
Quoted 25 lines
Documentation/process/submitting-patches.rst +0 -11
index cc6a1f73d7f2..b74c38aa9770 100644
@@ -634,16 +634,6 @@ bugzilla.kernel.org is a public place in this sense, but email addresses
 used there are private; so do not expose them in tags, unless the person
 used them in earlier contributions.
 
-Using Assisted-by:
-------------------
-
-If you used any sort of advanced coding tool in the creation of your patch,
-you need to acknowledge that use by adding an Assisted-by tag.  Failure to
-do so may impede the acceptance of your work.  Please see
-Documentation/process/coding-assistants.rst for details regarding the
-acknowledgment of coding assistants.
-
-
 .. _the_canonical_patch_format:
 
 The canonical patch format
 
---

base-commit: 665159e246749578d4e4bfe106ee3b74edcdab18
change-id: 20260702-aidoc-7e18f221d63a

Signature · 1 line
Jeff Layton <[email protected]>
#4 lore raw
Quoted 68 lines · Jeff Layton
Quoted 57 lines · Laurent Pinchart
Quoted 16 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

2/ It basically serves as free advertising for proprietary LLM companies.

3/ It's not clear why we want to collect this info in the first place.

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

Quoted 8 lines

---
Christian had proposed watering down the LLM attribution, but I think it's not productive to try and track this until we have a clearer sense of what we want to do with this information and how to make it more reliable.

---
 Documentation/process/coding-assistants.rst  | 22 ----------------------
 Documentation/process/submitting-patches.rst | 10 ----------

You're missing scripts/checkpatch.pl.

Quoted 29 lines
 2 files changed, 32 deletions(-)
Documentation/process/coding-assistants.rst +0 -19
index 899f4459c52d..c4cc0917fc92 100644
@@ -35,25 +35,3 @@ is responsible for:
 * Ensuring compliance with licensing requirements
 * Adding their own Signed-off-by tag to certify the DCO
 * Taking full responsibility for the contribution
-
-Attribution
-===========
-
-When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
-
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
-
-Where:
-
-* ``AGENT_NAME`` is the name of the AI tool or framework
-* ``MODEL_VERSION`` is the specific model version used
-* ``[TOOL1] [TOOL2]`` are optional specialized analysis tools used
-  (e.g., coccinelle, sparse, smatch, clang-tidy)
-
-Basic development tools (git, gcc, make, editors) should not be listed.
-
Quoted 9 lines

Documentation/process/coding-assistants.rst is responsible for: · line 59

Quoted 3 lines
Documentation/process/coding-assistants.rst +0 -3
-Example::
-
-  Assisted-by: Claude:claude-3-opus coccinelle sparse

Documentation/process/generated-content.rst still clearly requests contributors to disclose usage of code-generation tools. Even if we drop the Assisted-by tag, I think it is useful to mention the requirement here (probably including a link to generated-content.rst).

Quoted 2 lines

Fair point. Maybe we should keep the Attribution section then, and give it some brief text like:

I really do not understand how requiring disclosure but not tags isn't a contradictory position?

If disclosure via tags is worthless because people don't always do it, how is disclosure using rules not also worthless?

Tags seem like the clearest possible means of expressing this so now I'm confused as to the point being argued here...

Quoted 5 lines

"Contributions developed with LLM help should also follow the guidelines outlined in the generated-content document."

With a link to generated-content.rst?

Again I proposed to send this patch alrady but oppose your one, so how should we move forwards?

If the patch will land over my objections then fine I can save some time, but if it's in question, maybe I can just do that bit separately?...

Quoted 22 lines
Quoted 22 lines
Quoted 22 lines
Documentation/process/submitting-patches.rst +0 -10
index cc6a1f73d7f2..b74c38aa9770 100644
@@ -634,16 +634,6 @@ bugzilla.kernel.org is a public place in this sense, but email addresses
 used there are private; so do not expose them in tags, unless the person
 used them in earlier contributions.
 
-Using Assisted-by:
-------------------
-
-If you used any sort of advanced coding tool in the creation of your patch,
-you need to acknowledge that use by adding an Assisted-by tag.  Failure to
-do so may impede the acceptance of your work.  Please see
-Documentation/process/coding-assistants.rst for details regarding the
-acknowledgment of coding assistants.
-
-
 .. _the_canonical_patch_format:
 
Quoted 8 lines
Quoted 5 lines
Quoted 5 lines
Documentation/process/submitting-patches.rst +0 -1
 The canonical patch format
 
---

base-commit: 665159e246749578d4e4bfe106ee3b74edcdab18
change-id: 20260702-aidoc-7e18f221d63a

--
Jeff Layton <[email protected]>

Thanks, Lorenzo

#5 lore raw
In reply to Jeff Layton

I'm a little surprised I'm cc'd on this :) I'm not entirely sure if my pushing back on this is going to mean anything but I suppose here goes nothing.

Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Does something not being universally followed therefore make it worthless?

You really have to explain that, because this is literally true of any rule whatsoever we might have in the kernel, should we drop all of them?

I think you should replace this with a cogent argument such as that you feel it is not being used in the _majority_ of cases or is very rarely used, and that value is in your view not there.

Quoted 2 lines

2/ It basically serves as free advertising for proprietary LLM companies.

I agree with this point, we should drop the model.

Quoted 2 lines

3/ It's not clear why we want to collect this info in the first place.

Well I made arguments on the other thread, but to repeat:

- It makes it easier to engage with people when they do ack it.

- It makes it far quicker to be able to do so.

- There's a barrier to mentioning an LLM if it's not provided - people can get upset, or it can cause issues to raise it as a concern.

- Even if it's only there in _some_ cases, it makes _those_ cases easier to deal with.

- It provides some (incomplete) data that might make it easier to deal with bug-causing patches.

- It provides some (incomplete) data on bug rates with/without LLMs.

I do agree they're far far less useful when there's not some indication of how much of the patch was LLM-generated.

Their usefulness is obviously deeply far from perfect, but not zero.

Quoted 4 lines

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

I feel there are purposes. Perhaps the argument is stronger for having the tags on submissions rather than actually in-tree, however.

Quoted 2 lines

I think simplifying by requiring only 'LLM' and adding a line simply suggesting a comment giving a sense of how much was used, and then perhaps a paragraph saying

Subsystems may vary in their acceptance of this patch blah blah blah

Could be a compromise?

Quoted 5 lines

---
Christian had proposed watering down the LLM attribution, but I think it's not productive to try and track this until we have a clearer sense of what we want to do with this information and how to make it more reliable.

Well he's moved now to stridently saying we should do what you're doing here so presumably you're now in alignment.

Quoted 61 lines
---
 Documentation/process/coding-assistants.rst  | 22 ----------------------
 Documentation/process/submitting-patches.rst | 10 ----------
 2 files changed, 32 deletions(-)
Documentation/process/coding-assistants.rst +0 -22
index 899f4459c52d..c4cc0917fc92 100644
@@ -35,25 +35,3 @@ is responsible for:
 * Ensuring compliance with licensing requirements
 * Adding their own Signed-off-by tag to certify the DCO
 * Taking full responsibility for the contribution
-
-Attribution
-===========
-
-When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
-
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
-
-Where:
-
-* ``AGENT_NAME`` is the name of the AI tool or framework
-* ``MODEL_VERSION`` is the specific model version used
-* ``[TOOL1] [TOOL2]`` are optional specialized analysis tools used
-  (e.g., coccinelle, sparse, smatch, clang-tidy)
-
-Basic development tools (git, gcc, make, editors) should not be listed.
-
-Example::
-
-  Assisted-by: Claude:claude-3-opus coccinelle sparse
Documentation/process/submitting-patches.rst +0 -11
index cc6a1f73d7f2..b74c38aa9770 100644
@@ -634,16 +634,6 @@ bugzilla.kernel.org is a public place in this sense, but email addresses
 used there are private; so do not expose them in tags, unless the person
 used them in earlier contributions.
 
-Using Assisted-by:
-------------------
-
-If you used any sort of advanced coding tool in the creation of your patch,
-you need to acknowledge that use by adding an Assisted-by tag.  Failure to
-do so may impede the acceptance of your work.  Please see
-Documentation/process/coding-assistants.rst for details regarding the
-acknowledgment of coding assistants.
-
-
 .. _the_canonical_patch_format:
 
 The canonical patch format
 
---

base-commit: 665159e246749578d4e4bfe106ee3b74edcdab18
change-id: 20260702-aidoc-7e18f221d63a

Quoted 4 lines

Best regards,
--
Jeff Layton <[email protected]>

Thanks, Lorenzo

#6 lore raw

Hi Lorenzo,

Quoted 31 lines · Lorenzo Stoakes

I'm a little surprised I'm cc'd on this :) I'm not entirely sure if my pushing back on this is going to mean anything but I suppose here goes nothing.

Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Does something not being universally followed therefore make it worthless?

You really have to explain that, because this is literally true of any rule whatsoever we might have in the kernel, should we drop all of them?

I think you should replace this with a cogent argument such as that you feel it is not being used in the _majority_ of cases or is very rarely used, and that value is in your view not there.

Quoted 2 lines

2/ It basically serves as free advertising for proprietary LLM companies.

I agree with this point, we should drop the model.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

Well I made arguments on the other thread, but to repeat:

- It makes it easier to engage with people when they do ack it.

Quoted 7 lines

- It makes it far quicker to be able to do so.

- There's a barrier to mentioning an LLM if it's not provided - people can get upset, or it can cause issues to raise it as a concern.

- Even if it's only there in _some_ cases, it makes _those_ cases easier to deal with.

As far as I understand, all the above arguments would also be addressed with either a free-formed mention of LLM usage, or a formal "non-advertising" tag that is not merged in the kernel history, right ?

Quoted 4 lines

- It provides some (incomplete) data that might make it easier to deal with bug-causing patches.

- It provides some (incomplete) data on bug rates with/without LLMs.

For those two I suppose a machine-parseable tag in the git history could improve things slightly compared to information provided in the patch submission that would not get recorded in the history.

Is this why you have a preference for a formal tag compared to a free-formed mention ?

Quoted 11 lines

I do agree they're far far less useful when there's not some indication of how much of the patch was LLM-generated.

Their usefulness is obviously deeply far from perfect, but not zero.

Quoted 3 lines

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

I feel there are purposes. Perhaps the argument is stronger for having the tags on submissions rather than actually in-tree, however.

I really think we need to have the information at submission. I think I have a slight preference for not recording it in the kernel tree, but only slight.

Quoted 78 lines
Quoted 1 line

I think simplifying by requiring only 'LLM' and adding a line simply suggesting a comment giving a sense of how much was used, and then perhaps a paragraph saying

Subsystems may vary in their acceptance of this patch blah blah blah

Could be a compromise?

Quoted 5 lines

---
Christian had proposed watering down the LLM attribution, but I think it's not productive to try and track this until we have a clearer sense of what we want to do with this information and how to make it more reliable.

Well he's moved now to stridently saying we should do what you're doing here so presumably you're now in alignment.

Quoted 60 lines
---
 Documentation/process/coding-assistants.rst  | 22 ----------------------
 Documentation/process/submitting-patches.rst | 10 ----------
 2 files changed, 32 deletions(-)
Documentation/process/coding-assistants.rst +0 -22
index 899f4459c52d..c4cc0917fc92 100644
@@ -35,25 +35,3 @@ is responsible for:
 * Ensuring compliance with licensing requirements
 * Adding their own Signed-off-by tag to certify the DCO
 * Taking full responsibility for the contribution
-
-Attribution
-===========
-
-When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
-
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
-
-Where:
-
-* ``AGENT_NAME`` is the name of the AI tool or framework
-* ``MODEL_VERSION`` is the specific model version used
-* ``[TOOL1] [TOOL2]`` are optional specialized analysis tools used
-  (e.g., coccinelle, sparse, smatch, clang-tidy)
-
-Basic development tools (git, gcc, make, editors) should not be listed.
-
-Example::
-
-  Assisted-by: Claude:claude-3-opus coccinelle sparse
Documentation/process/submitting-patches.rst +0 -11
index cc6a1f73d7f2..b74c38aa9770 100644
@@ -634,16 +634,6 @@ bugzilla.kernel.org is a public place in this sense, but email addresses
 used there are private; so do not expose them in tags, unless the person
 used them in earlier contributions.
 
-Using Assisted-by:
-------------------
-
-If you used any sort of advanced coding tool in the creation of your patch,
-you need to acknowledge that use by adding an Assisted-by tag.  Failure to
-do so may impede the acceptance of your work.  Please see
-Documentation/process/coding-assistants.rst for details regarding the
-acknowledgment of coding assistants.
-
-
 .. _the_canonical_patch_format:
 
 The canonical patch format
 
---

base-commit: 665159e246749578d4e4bfe106ee3b74edcdab18
change-id: 20260702-aidoc-7e18f221d63a

Signature · 3 lines
Regards,

Laurent Pinchart
#7 lore raw
Quoted 39 lines · Laurent Pinchart

Hi Lorenzo,

Quoted 36 lines · Lorenzo Stoakes

I'm a little surprised I'm cc'd on this :) I'm not entirely sure if my pushing back on this is going to mean anything but I suppose here goes nothing.

Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Does something not being universally followed therefore make it worthless?

You really have to explain that, because this is literally true of any rule whatsoever we might have in the kernel, should we drop all of them?

I think you should replace this with a cogent argument such as that you feel it is not being used in the _majority_ of cases or is very rarely used, and that value is in your view not there.

Quoted 2 lines

2/ It basically serves as free advertising for proprietary LLM companies.

I agree with this point, we should drop the model.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

Well I made arguments on the other thread, but to repeat:

- It makes it easier to engage with people when they do ack it.

- It makes it far quicker to be able to do so.

- There's a barrier to mentioning an LLM if it's not provided - people can get upset, or it can cause issues to raise it as a concern.

Quoted 6 lines
Quoted 2 lines

- Even if it's only there in _some_ cases, it makes _those_ cases easier to deal with.

As far as I understand, all the above arguments would also be addressed with either a free-formed mention of LLM usage, or a formal "non-advertising" tag that is not merged in the kernel history, right ?

Well I prefer a tag for reasons below. I'm fine with non-advertising yeah.

Quoted 9 lines
Quoted 5 lines

- It provides some (incomplete) data that might make it easier to deal with bug-causing patches.

- It provides some (incomplete) data on bug rates with/without LLMs.

For those two I suppose a machine-parseable tag in the git history could improve things slightly compared to information provided in the patch submission that would not get recorded in the history.

Yeah, again I could live without it being in the tree. I think there's some advantages though.

See below for tag argument.

Quoted 13 lines

Is this why you have a preference for a formal tag compared to a free-formed mention ?

Quoted 9 lines

I do agree they're far far less useful when there's not some indication of how much of the patch was LLM-generated.

Their usefulness is obviously deeply far from perfect, but not zero.

Quoted 3 lines

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

Quoted 6 lines
Quoted 2 lines

I feel there are purposes. Perhaps the argument is stronger for having the tags on submissions rather than actually in-tree, however.

I really think we need to have the information at submission. I think I have a slight preference for not recording it in the kernel tree, but only slight.

It seems we are more in agreement than it seems :)

I prefer a tag as it's clear cut and easily greppable and easily noticeable.

If it's words it can be vague or be unclear or inconsistent and you might miss etc.

So I have a strong preference for a tag, would love to add details about how much LLM done (that'll be vague obviously + have grey lines).

I agree gettig rid of model is worthwhile but I'm not bothered if we don't do that.

Thanks, Lorenzo

#8 lore raw
Quoted 75 lines · Lorenzo Stoakes
Quoted 39 lines · Laurent Pinchart

Hi Lorenzo,

Quoted 36 lines · Lorenzo Stoakes

I'm a little surprised I'm cc'd on this :) I'm not entirely sure if my pushing back on this is going to mean anything but I suppose here goes nothing.

Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Does something not being universally followed therefore make it worthless?

You really have to explain that, because this is literally true of any rule whatsoever we might have in the kernel, should we drop all of them?

I think you should replace this with a cogent argument such as that you feel it is not being used in the _majority_ of cases or is very rarely used, and that value is in your view not there.

Quoted 2 lines

2/ It basically serves as free advertising for proprietary LLM companies.

I agree with this point, we should drop the model.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

Well I made arguments on the other thread, but to repeat:

- It makes it easier to engage with people when they do ack it.

- It makes it far quicker to be able to do so.

- There's a barrier to mentioning an LLM if it's not provided - people can get upset, or it can cause issues to raise it as a concern.

Quoted 6 lines
Quoted 2 lines

- Even if it's only there in _some_ cases, it makes _those_ cases easier to deal with.

As far as I understand, all the above arguments would also be addressed with either a free-formed mention of LLM usage, or a formal "non-advertising" tag that is not merged in the kernel history, right ?

Well I prefer a tag for reasons below. I'm fine with non-advertising yeah.

Quoted 8 lines
Quoted 4 lines

- It provides some (incomplete) data that might make it easier to deal with bug-causing patches.

- It provides some (incomplete) data on bug rates with/without LLMs.

For those two I suppose a machine-parseable tag in the git history could improve things slightly compared to information provided in the patch submission that would not get recorded in the history.

Yeah, again I could live without it being in the tree. I think there's some advantages though.

See below for tag argument.

Quoted 12 lines

Is this why you have a preference for a formal tag compared to a free-formed mention ?

Quoted 9 lines

I do agree they're far far less useful when there's not some indication of how much of the patch was LLM-generated.

Their usefulness is obviously deeply far from perfect, but not zero.

Quoted 3 lines

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

I feel there are purposes. Perhaps the argument is stronger for having the tags on submissions rather than actually in-tree, however.

I really think we need to have the information at submission. I think I have a slight preference for not recording it in the kernel tree, but only slight.

It seems we are more in agreement than it seems :)

I knew we were :-)

Quoted 7 lines

I prefer a tag as it's clear cut and easily greppable and easily noticeable.

If it's words it can be vague or be unclear or inconsistent and you might miss etc.

So I have a strong preference for a tag, would love to add details about how much LLM done (that'll be vague obviously + have grey lines).

I have a slight preference for not recording that in the git history, but I can live with either.

Quoted 2 lines

I agree gettig rid of model is worthwhile but I'm not bothered if we don't do that.

There I have a very strong preference for stopping free advertising.

Signature · 3 lines
Regards,

Laurent Pinchart
#9 lore raw
In reply to Jeff Layton
Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Then let's move to get people to follow it.

Quoted 1 line

2/ It basically serves as free advertising for proprietary LLM companies.

Who cares, make up a name, all I want is the "signal" that someone is using a LLM so that I can review it as-such. And if I think someone is not reporting that, I can ask for them to properly attribute it and if they lie, well, that's on them.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

We want to know if a LLM is being used.

Quoted 3 lines

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

No, please do not do this. It's useful already for many patches in my subsystems, and is only going to be used more in the future.

thanks,

greg k-h

maintainer
#10 lore raw
Quoted 9 lines · Greg KH
Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Then let's move to get people to follow it.

FWIW, in my experience, simple ignorance of the rules is one of the biggest reasons why so many people fail to follow them. When I've asked docs contributors about the source of their work, they tend to come clean, apologize, and add the tags going forward.

jon

#11 lore raw
Quoted 10 lines · Jonathan Corbet
Quoted 8 lines · Greg KH
Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Quoted 6 lines
Quoted 1 line

Then let's move to get people to follow it.

FWIW, in my experience, simple ignorance of the rules is one of the biggest reasons why so many people fail to follow them. When I've asked docs contributors about the source of their work, they tend to come clean, apologize, and add the tags going forward.

I also feel that repeatedly establishing boundaries like this (in as civil and reasonable way as possible) can help establish the 'done thing'.

We will probably still experience a great deal of unattributed patches, and I think over time we'll have to figure out a way to deal with that, but there's certainly a cohort who can be reached, and it's beneficial to do so.

It also makes it easier to enforce with those who don't play nice - 'here are the rules, look over here, and look at these patches' etc.

With no rules at all it becomes far harder to enforce anything.

Quoted 2 lines

jon

Thanks, Lorenzo

#12 lore raw
In reply to Jonathan Corbet
Quoted 11 lines · Jonathan Corbet
Quoted 8 lines · Greg KH
Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Quoted 6 lines
Quoted 1 line

Then let's move to get people to follow it.

FWIW, in my experience, simple ignorance of the rules is one of the biggest reasons why so many people fail to follow them. When I've asked docs contributors about the source of their work, they tend to come clean, apologize, and add the tags going forward.

It would make sense IMHO to address the ignorance in the most expedient manner, namely by telling the *LLM itself* to add this tag when it generates the commit message and/or reviews the code.

This could be achieved by adding a statement in "AGENTS.md" in the root of the source tree to this effect, or telling the agent to reference and follow rules in Documentation/process/coding-assistants.rst regarding the Assisted-by: tag.

Cheers, Andreas

#13 lore raw
Quoted 7 lines · Andreas Dilger

It would make sense IMHO to address the ignorance in the most expedient manner, namely by telling the *LLM itself* to add this tag when it generates the commit message and/or reviews the code.

This could be achieved by adding a statement in "AGENTS.md" in the root of the source tree to this effect, or telling the agent to reference and follow rules in Documentation/process/coding-assistants.rst regarding the Assisted-by: tag.

This presupposes a particular workflow where the developer allows the LLM to generate git commits. I don't do that. I'll ask the LLM to modify the code, but then I'll generally fix it up --- very often by hand because I can edit the sources than entering a prompt and waiting for the LLM to figure out how to make the change. I'll often run the test cases myself, since if it takes 24 hours of VM time, and 2 hours of wall clock time, I'm going to kick off the regression tests myself, and verify the test results.

And I almost always write the commit description myself. So when I say, "Assisted-by" it literally is "Assisted by". It is not "vibe coded where the LLM generates thousands of lines of code that the human being doesn't understand before sending ta pull request." And that's probably why you'll see people asserting that LLM generated code can't possibly be accepted due to copyright reasons. There's a big difference between asking an LLM to modify already existing code to add a feature, or to fix a bug, and to vibe code a new OS from scratch. One is far less likely to be a copyright violation than the other.

- Ted

#14 lore raw
Quoted 33 lines

Op 03-07-2026 04:57 CEST schreef Theodore Tso <[email protected]>:

Quoted 7 lines · Andreas Dilger

It would make sense IMHO to address the ignorance in the most expedient manner, namely by telling the *LLM itself* to add this tag when it generates the commit message and/or reviews the code.

This could be achieved by adding a statement in "AGENTS.md" in the root of the source tree to this effect, or telling the agent to reference and follow rules in Documentation/process/coding-assistants.rst regarding the Assisted-by: tag.

This presupposes a particular workflow where the developer allows the LLM to generate git commits. I don't do that. I'll ask the LLM to modify the code, but then I'll generally fix it up --- very often by hand because I can edit the sources than entering a prompt and waiting for the LLM to figure out how to make the change. I'll often run the test cases myself, since if it takes 24 hours of VM time, and 2 hours of wall clock time, I'm going to kick off the regression tests myself, and verify the test results.

And I almost always write the commit description myself. So when I say, "Assisted-by" it literally is "Assisted by". It is not "vibe coded where the LLM generates thousands of lines of code that the human being doesn't understand before sending ta pull request." And that's probably why you'll see people asserting that LLM generated code can't possibly be accepted due to copyright reasons. There's a big difference between asking an LLM to modify already existing code to add a feature, or to fix a bug, and to vibe code a new OS from scratch. One is far less likely to be a copyright violation than the other.

Quoted 1 line

- Ted

Just curious if you know about any quantifiable insight into this. I do know that the issues also plays up in smaller requests. I once asked an LLM to give me boilerplate for a FUSE fs, and it took almost verbatim the example hello.c from the repo (which I only recognized because I had looked at that code) but stripped the copyright info. And this is only like 100 lines of code or so.

Best,
Jori.

#15 lore raw
In reply to Theodore Tso
Quoted 10 lines · Theodore Tso
Quoted 7 lines · Andreas Dilger

It would make sense IMHO to address the ignorance in the most expedient manner, namely by telling the *LLM itself* to add this tag when it generates the commit message and/or reviews the code.

This could be achieved by adding a statement in "AGENTS.md" in the root of the source tree to this effect, or telling the agent to reference and follow rules in Documentation/process/coding-assistants.rst regarding the Assisted-by: tag.

Quoted 8 lines

This presupposes a particular workflow where the developer allows the LLM to generate git commits. I don't do that. I'll ask the LLM to modify the code, but then I'll generally fix it up --- very often by hand because I can edit the sources than entering a prompt and waiting for the LLM to figure out how to make the change. I'll often run the test cases myself, since if it takes 24 hours of VM time, and 2 hours of wall clock time, I'm going to kick off the regression tests myself, and verify the test results.

Sure, and that holds true for many kernel developers, myself included. I don't think you or any of the senior maintainers are the audience.

The point of my comment is that this is more likely to help contributors who do *not* know/follow standard development practices and are more likely to contribute AI-coded patches, which is AFAICS the main issue in this discussion.

Also, the effort of doing so is very minimal and has no real downside. Having the AI contribution policy clearly spelled out where an AI agent can find it can only have a positive effect, even if not always followed.

Cheers, Andreas

Quoted 11 lines

And I almost always write the commit description myself. So when I say, "Assisted-by" it literally is "Assisted by". It is not "vibe coded where the LLM generates thousands of lines of code that the human being doesn't understand before sending ta pull request." And that's probably why you'll see people asserting that LLM generated code can't possibly be accepted due to copyright reasons. There's a big difference between asking an LLM to modify already existing code to add a feature, or to fix a bug, and to vibe code a new OS from scratch. One is far less likely to be a copyright violation than the other.

Quoted 1 line

- Ted

Cheers, Andreas

#16 lore raw
In reply to Greg KH
Quoted 25 lines · Greg KH
Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Then let's move to get people to follow it.

Quoted 1 line

2/ It basically serves as free advertising for proprietary LLM companies.

Who cares, make up a name, all I want is the "signal" that someone is using a LLM so that I can review it as-such. And if I think someone is not reporting that, I can ask for them to properly attribute it and if they lie, well, that's on them.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

We want to know if a LLM is being used.

Quoted 3 lines

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

Quoted 6 lines

No, please do not do this. It's useful already for many patches in my subsystems, and is only going to be used more in the future.

thanks,

greg k-h

Entirely my view and experience. The tags are proactively useful, even if flawed and (possibly very) incomplete.

I am totally fine with getting rid of the model however, and I do think it'd be useful to add a small paragraph suggesting that people add a comment indicating _how much_ of the patch was LLM-generated, even if it's vague and fuzzy.

Some information > no information.

Thanks, Lorenzo

#17 lore raw
In reply to Greg KH
Quoted 29 lines

Op 02-07-2026 17:07 CEST schreef Greg KH <[email protected]>:

Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Then let's move to get people to follow it.

Quoted 1 line

2/ It basically serves as free advertising for proprietary LLM companies.

Who cares, make up a name, all I want is the "signal" that someone is using a LLM so that I can review it as-such. And if I think someone is not reporting that, I can ask for them to properly attribute it and if they lie, well, that's on them.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

We want to know if a LLM is being used.

Quoted 3 lines

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

Quoted 6 lines

No, please do not do this. It's useful already for many patches in my subsystems, and is only going to be used more in the future.

thanks,

greg k-h

Why not:

Keep the tag like David suggested as:

Assisted-by LLM # automated removal of useless blabla

I am indifferent to whether this should be a tag or just below the --- .

And add something like the systemd guidelines that Christian linked to to the AI Coding Assistants doc page. This provides much more useful guidelines than what is currently there. I triggered this whole discussion by reading the page and adding the model to some clean-up commits, so some people do read it...

Best,
Jori.

#18 lore raw
Quoted 40 lines · Jori Koolstra
Quoted 29 lines

Op 02-07-2026 17:07 CEST schreef Greg KH <[email protected]>:

Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Then let's move to get people to follow it.

Quoted 1 line

2/ It basically serves as free advertising for proprietary LLM companies.

Who cares, make up a name, all I want is the "signal" that someone is using a LLM so that I can review it as-such. And if I think someone is not reporting that, I can ask for them to properly attribute it and if they lie, well, that's on them.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

We want to know if a LLM is being used.

Quoted 3 lines

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

Quoted 6 lines

No, please do not do this. It's useful already for many patches in my subsystems, and is only going to be used more in the future.

thanks,

greg k-h

Why not:

Keep the tag like David suggested as:

Quoted 8 lines
Assisted-by LLM # automated removal of useless blabla

I am indifferent to whether this should be a tag or just below the --- .

And add something like the systemd guidelines that Christian linked to to the AI Coding Assistants doc page. This provides much more useful guidelines than what is currently there. I triggered this whole discussion by reading the page and adding the model to some clean-up commits, so some people do read it...

Yup am fine with all that!

Quoted 3 lines

Best,
Jori.

Thanks, Lorenzo

#19 lore raw
In reply to Greg KH
Quoted 9 lines · Greg KH
Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Then let's move to get people to follow it.

Has there been any discussion regarding *how* to do so ? Jon mentioned that in many cases the failure to disclose is simply due to ignorance. That could be addressed by more advertising (not entirely sure where though, documentation isn't always read), or restructuring of coding-assistants.rst (as far as I understand, the file is supposed to be consumed by agents, so it should be reasonable to expect the LLM output to contain a tag).

There are also cases where submitters refuse to comply with the rules (as in [1] apparently). For those cases, a clear message that maintainers will be fully supported by the community when they push back could help.

[1] https://lore.kernel.org/linux-mm/[email protected]/

Quoted 22 lines
Quoted 1 line

2/ It basically serves as free advertising for proprietary LLM companies.

Who cares, make up a name, all I want is the "signal" that someone is using a LLM so that I can review it as-such. And if I think someone is not reporting that, I can ask for them to properly attribute it and if they lie, well, that's on them.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

We want to know if a LLM is being used.

Quoted 3 lines

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

No, please do not do this. It's useful already for many patches in my subsystems, and is only going to be used more in the future.

thanks,

greg k-h

Signature · 3 lines
Regards,

Laurent Pinchart
#20 lore raw
In reply to Greg KH
Quoted 12 lines · Greg KH
Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Then let's move to get people to follow it.

Quoted 1 line

2/ It basically serves as free advertising for proprietary LLM companies.

Quoted 9 lines

Who cares, make up a name, all I want is the "signal" that someone is using a LLM so that I can review it as-such. And if I think someone is not reporting that, I can ask for them to properly attribute it and if they lie, well, that's on them.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

We want to know if a LLM is being used.

But why? What do you intend to do with this information?

Do you mean to use it as an indicator that the patch should receive "extra" review (or maybe that it should be ignored)? Do you mean to use it to generate some sort of statistics at a later time?

If we want to collect some sort of structured data like this, it would be good to have some idea of how it will be used. That might inform the format and content of what we're collecting.

Quoted 7 lines
Quoted 3 lines

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

No, please do not do this. It's useful already for many patches in my subsystems, and is only going to be used more in the future.

I should have sent this as an RFC patch, but hopefully everyone is treating that way anyway. To be clear, I don't have a strong preference on whether we keep Assisted-by: or not, but I'd like to better understand the intended uses of it if we do intend to keep it.

Signature · 1 line
Jeff Layton <[email protected]>
#21 lore raw
Quoted 20 lines · Jeff Layton
Quoted 19 lines · Greg KH
Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Then let's move to get people to follow it.

Quoted 1 line

2/ It basically serves as free advertising for proprietary LLM companies.

Who cares, make up a name, all I want is the "signal" that someone is using a LLM so that I can review it as-such. And if I think someone is not reporting that, I can ask for them to properly attribute it and if they lie, well, that's on them.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

Quoted 7 lines
Quoted 1 line

We want to know if a LLM is being used.

But why? What do you intend to do with this information?

Do you mean to use it as an indicator that the patch should receive "extra" review (or maybe that it should be ignored)? Do you mean to use it to generate some sort of statistics at a later time?

I use the information to decide how to review the patch, and what level of priority to give it. For that usage I don't need a tag, but I need the information in some human-readable form at patch submission time.

Quoted 15 lines

If we want to collect some sort of structured data like this, it would be good to have some idea of how it will be used. That might inform the format and content of what we're collecting.

Quoted 6 lines
Quoted 3 lines

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

No, please do not do this. It's useful already for many patches in my subsystems, and is only going to be used more in the future.

I should have sent this as an RFC patch, but hopefully everyone is treating that way anyway. To be clear, I don't have a strong preference on whether we keep Assisted-by: or not, but I'd like to better understand the intended uses of it if we do intend to keep it.

Signature · 3 lines
Regards,

Laurent Pinchart
#22 lore raw
Quoted 25 lines · Laurent Pinchart
Quoted 24 lines · Jeff Layton
Quoted 12 lines · Greg KH
Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Then let's move to get people to follow it.

Quoted 1 line

2/ It basically serves as free advertising for proprietary LLM companies.

Quoted 8 lines

Who cares, make up a name, all I want is the "signal" that someone is using a LLM so that I can review it as-such. And if I think someone is not reporting that, I can ask for them to properly attribute it and if they lie, well, that's on them.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

We want to know if a LLM is being used.

But why? What do you intend to do with this information?

Quoted 7 lines
Quoted 3 lines

Do you mean to use it as an indicator that the patch should receive "extra" review (or maybe that it should be ignored)? Do you mean to use it to generate some sort of statistics at a later time?

I use the information to decide how to review the patch, and what level of priority to give it. For that usage I don't need a tag, but I need the information in some human-readable form at patch submission time.

Same here. I don't care about stats, I care about "how do I review this patch" and this gives me that signal that I need if faced with a llm-helped patch.

So it needs to stay please.

thanks,

greg k-h

#23 lore raw
Quoted 34 lines · Greg KH
Quoted 25 lines · Laurent Pinchart
Quoted 24 lines · Jeff Layton
Quoted 12 lines · Greg KH
Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Then let's move to get people to follow it.

Quoted 1 line

2/ It basically serves as free advertising for proprietary LLM companies.

Quoted 8 lines

Who cares, make up a name, all I want is the "signal" that someone is using a LLM so that I can review it as-such. And if I think someone is not reporting that, I can ask for them to properly attribute it and if they lie, well, that's on them.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

We want to know if a LLM is being used.

But why? What do you intend to do with this information?

Quoted 7 lines
Quoted 3 lines

Do you mean to use it as an indicator that the patch should receive "extra" review (or maybe that it should be ignored)? Do you mean to use it to generate some sort of statistics at a later time?

I use the information to decide how to review the patch, and what level of priority to give it. For that usage I don't need a tag, but I need the information in some human-readable form at patch submission time.

Quoted 5 lines

Same here. I don't care about stats, I care about "how do I review this patch" and this gives me that signal that I need if faced with a llm-helped patch.

So it needs to stay please.

There's the free advertising issue we still need to address. Would the proposed "Assisted-by: LLM" tag to replace product names be an acceptable option for you ?

Signature · 3 lines
Regards,

Laurent Pinchart
#24 lore raw
Quoted 39 lines · Laurent Pinchart
Quoted 38 lines · Greg KH
Quoted 25 lines · Laurent Pinchart
Quoted 24 lines · Jeff Layton
Quoted 12 lines · Greg KH
Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Then let's move to get people to follow it.

Quoted 1 line

2/ It basically serves as free advertising for proprietary LLM companies.

Quoted 8 lines

Who cares, make up a name, all I want is the "signal" that someone is using a LLM so that I can review it as-such. And if I think someone is not reporting that, I can ask for them to properly attribute it and if they lie, well, that's on them.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

We want to know if a LLM is being used.

But why? What do you intend to do with this information?

Quoted 7 lines
Quoted 3 lines

Do you mean to use it as an indicator that the patch should receive "extra" review (or maybe that it should be ignored)? Do you mean to use it to generate some sort of statistics at a later time?

I use the information to decide how to review the patch, and what level of priority to give it. For that usage I don't need a tag, but I need the information in some human-readable form at patch submission time.

Same here. I don't care about stats, I care about "how do I review this patch" and this gives me that signal that I need if faced with a llm-helped patch.

Quoted 5 lines
Quoted 1 line

So it needs to stay please.

There's the free advertising issue we still need to address. Would the proposed "Assisted-by: LLM" tag to replace product names be an acceptable option for you ?

I really don't care what the tag says, I just want to see the tag stay to show that this needs to be evaluated differently than if it was made only by a person. Some people might want to "show off" their model names, while other companies I know refuse to allow the name to be shown, so it really doesn't hurt anyone to keep it as-is.

thanks,

greg k-h

#25 lore raw
Quoted 7 lines · Greg KH
Quoted 6 lines · Laurent Pinchart
Quoted 5 lines · Greg KH

Same here. I don't care about stats, I care about "how do I review this patch" and this gives me that signal that I need if faced with a llm-helped patch.

Quoted 7 lines
Quoted 5 lines
Quoted 1 line

So it needs to stay please.

There's the free advertising issue we still need to address. Would the proposed "Assisted-by: LLM" tag to replace product names be an acceptable option for you ?

I really don't care what the tag says,

Well, while you don't care, the current policy implies that:

"

Assisted-by AGENT_NAME:MODEL_VERSION

...
``MODEL_VERSION`` is the specific model version used "

So clarifying there that a "Assisted-by: LLM" is also good enough would make sense if we agree on that.

I'll also note, that the "*should* include an Assisted-by tag" is pretty weak, as someone who worked on specs before.

My kids should clean up their room; doesn't work.

Really, if we want this to be useful in practice I only see the following way:

1) Document that it's a *must*, an document which scenarios we care about.

2) Encourage people to specify *how* AI was used.

Assisted-by LLM # improve the patch description

Now, it will be hard for all susbsytems to agree on that, so I would hope that Linus would just tell us what we should do :)

Signature · 3 lines
Cheers,

David
#26 lore raw
Quoted 18 lines · David Hildenbrand (Arm)
Quoted 7 lines · Greg KH
Quoted 6 lines · Laurent Pinchart
Quoted 5 lines · Greg KH

Same here. I don't care about stats, I care about "how do I review this patch" and this gives me that signal that I need if faced with a llm-helped patch.

Quoted 7 lines
Quoted 5 lines
Quoted 1 line

So it needs to stay please.

There's the free advertising issue we still need to address. Would the proposed "Assisted-by: LLM" tag to replace product names be an acceptable option for you ?

I really don't care what the tag says,

Well, while you don't care, the current policy implies that:

Quoted 8 lines

"

Assisted-by AGENT_NAME:MODEL_VERSION

...
``MODEL_VERSION`` is the specific model version used "

So clarifying there that a "Assisted-by: LLM" is also good enough would make sense if we agree on that.

If we all agree on that, I'm saying that _I_ don't care, but others might.

For me, the info after "Assisted-by:" doesn't matter, it's the first part.

Quoted 4 lines

I'll also note, that the "*should* include an Assisted-by tag" is pretty weak, as someone who worked on specs before.

My kids should clean up their room; doesn't work.

But, if you notice that your kids didn't clean up their room, and you had told them to, you can then talk to them about doing it properly based on what they are supposed to be doing.

That's the point here.

thanks,

greg k-h

#27 lore raw
Quoted 10 lines
Quoted 4 lines

I'll also note, that the "*should* include an Assisted-by tag" is pretty weak, as someone who worked on specs before.

My kids should clean up their room; doesn't work.

But, if you notice that your kids didn't clean up their room, and you had told them to, you can then talk to them about doing it properly based on what they are supposed to be doing.

That's the point here.

Well, they are smart. They will tell me that it was not a strict requirement.

So I'll have to rephrase it as "STOP ARGUING, CLEAN UP YOUR ROOM!!!" ;)

Signature · 3 lines
Cheers,

David
#28 lore raw
In reply to Greg KH
Quoted 5 lines · Greg KH
Quoted 2 lines

So clarifying there that a "Assisted-by: LLM" is also good enough would make sense if we agree on that.

If we all agree on that, I'm saying that _I_ don't care, but others might.

I agree that requiring the model name is not a good idea, because some developers might be using unreleased products where the name itself might not be public. So allowing the version number of the LLM to be elided would be a good change.

I'll also note that it's starting to be the case where tool is actually quite relevant. So it's not just about the LLM Model, but whether you're using Codex, Claude Code, OpenCode, or Shahiko. You can use the tool or the harness with different models.

So whether the tag is:

Assisted-By OpenCode:Gemini 3.5 Pro
Assisted-By Claude Code
Assisted-By LLM

I'd be fine with reviewing patches with any of the above. However, to me that's actually not the most interesting part. See below....

Quoted 2 lines

For me, the info after "Assisted-by:" doesn't matter, it's the first part.

For me, what I care most about is *how* the LLM was used. For example, if someone just used the output of Sashiko to fix a problem in their commit, that's one kind of "Assisted-by". If Sahsiko identifies a pre-existing bug, and the developer addes a patch to the patch series, or creates a new patch series, that's a different kind of "Assisted-by". A third kind of "Assisted-by" might include asking the LLM to create a Coccinelle semantic patches (because I'm not smart enough to create semantic patches).

And of course, all of this is quite different from the stereotypical "vibe coding" where the LLM generates thousands of lines of code which the human doesn't understand before sending the pull request. :-)

So for me, adding something after a #-sign comment explaning how the LLM was used would be very useful.

Quoted 5 lines
Quoted 1 line

My kids should clean up their room; doesn't work.

But, if you notice that your kids didn't clean up their room, and you had told them to, you can then talk to them about doing it properly based on what they are supposed to be doing.

Another metaphor might be that people shouldn't try to evade paying their taxes; merely passing a law saying they have to file correct tax returns doesn't guarantee this. But it was specifically the law against tax evasion is how US Law Enforcement arrested the Chicago mob boss Al Capone.

Just because some people might lie doesn't mean that we shouldn't bother to ask everyone to fill out tax returns.

Cheers,

- Ted

#29 lore raw
Quoted 51 lines · Theodore Tso
Quoted 5 lines · Greg KH
Quoted 2 lines

So clarifying there that a "Assisted-by: LLM" is also good enough would make sense if we agree on that.

If we all agree on that, I'm saying that _I_ don't care, but others might.

I agree that requiring the model name is not a good idea, because some developers might be using unreleased products where the name itself might not be public. So allowing the version number of the LLM to be elided would be a good change.

I'll also note that it's starting to be the case where tool is actually quite relevant. So it's not just about the LLM Model, but whether you're using Codex, Claude Code, OpenCode, or Shahiko. You can use the tool or the harness with different models.

So whether the tag is:

Assisted-By OpenCode:Gemini 3.5 Pro
Assisted-By Claude Code
Assisted-By LLM

I'd be fine with reviewing patches with any of the above. However, to me that's actually not the most interesting part. See below....

Quoted 2 lines

For me, the info after "Assisted-by:" doesn't matter, it's the first part.

For me, what I care most about is *how* the LLM was used. For example, if someone just used the output of Sashiko to fix a problem in their commit, that's one kind of "Assisted-by". If Sahsiko identifies a pre-existing bug, and the developer addes a patch to the patch series, or creates a new patch series, that's a different kind of "Assisted-by". A third kind of "Assisted-by" might include asking the LLM to create a Coccinelle semantic patches (because I'm not smart enough to create semantic patches).

And of course, all of this is quite different from the stereotypical "vibe coding" where the LLM generates thousands of lines of code which the human doesn't understand before sending the pull request. :-)

So for me, adding something after a #-sign comment explaning how the LLM was used would be very useful.

Quoted 5 lines
Quoted 1 line

My kids should clean up their room; doesn't work.

But, if you notice that your kids didn't clean up their room, and you had told them to, you can then talk to them about doing it properly based on what they are supposed to be doing.

Quoted 8 lines

Another metaphor might be that people shouldn't try to evade paying their taxes; merely passing a law saying they have to file correct tax returns doesn't guarantee this. But it was specifically the law against tax evasion is how US Law Enforcement arrested the Chicago mob boss Al Capone.

Just because some people might lie doesn't mean that we shouldn't bother to ask everyone to fill out tax returns.

I think we're digressing here. The original point, if I understood it carefully, is that "must" would convey a requirement better than "should", and I agree with that.

Signature · 3 lines
Regards,

Laurent Pinchart
#30 lore raw
In reply to Theodore Tso
Quoted 9 lines

Op 03-07-2026 13:42 CEST schreef Theodore Tso <[email protected]>:

Another metaphor might be that people shouldn't try to evade paying their taxes; merely passing a law saying they have to file correct tax returns doesn't guarantee this. But it was specifically the law against tax evasion is how US Law Enforcement arrested the Chicago mob boss Al Capone.

Quoted 6 lines

Just because some people might lie doesn't mean that we shouldn't bother to ask everyone to fill out tax returns.

Cheers,

- Ted

I think the point that some people who are more negative on keeping the tag in were trying to make is not that we expect everyone to follow the rules, but that having rules with no consequences are bad rules in general. And this is illustrated by your example.

But I see there is quite a bit of consensus on the general needs:

- it's more relevant to know how an LLM was used than that it was used (and the tag is just a quick indication of this)
- advertising for particular corporations does not really serve any purpose to the community, so just say "LLM"
- the tags may be used to prevent arguing
- the requirement to be open about LLM use (whether by tags or whatever) makes it easier for maintainers (if they feel the need) to de-prioritize patches by someone unwilling to abide by these rules whenever the maintainer perceives or suspects prior undeclared LLM use. This prevents an asymmetrical situation of LLM slop being produced much faster that what can be reviewed.

#31 lore raw
In reply to Theodore Tso
Quoted 17 lines · On Fri, 3 Jul 2026 07:42:35 -0400 "Theodore Tso" <[email protected]> wrote:
Quoted 5 lines · Greg KH
Quoted 2 lines

So clarifying there that a "Assisted-by: LLM" is also good enough would make sense if we agree on that.

If we all agree on that, I'm saying that _I_ don't care, but others might.

I agree that requiring the model name is not a good idea, because some developers might be using unreleased products where the name itself might not be public. So allowing the version number of the LLM to be elided would be a good change.

I'll also note that it's starting to be the case where tool is actually quite relevant. So it's not just about the LLM Model, but whether you're using Codex, Claude Code, OpenCode, or Shahiko. You can use the tool or the harness with different models.

Quoted 5 lines

So whether the tag is:

Assisted-By OpenCode:Gemini 3.5 Pro
Assisted-By Claude Code
Assisted-By LLM

I don't think an assisted-by for LLM actually helps.

I can't see any difference on adding it or doing something like:

	Assisted-By: smath
	Assisted-By: sparse
	Assisted-By: coverity
	Assisted-By: gcc
	...
	Assisted-By: my-favorite-web-search-engine

From my side, I don't care about whatever tool/process a developer has to produce a patch. All I want is a proper patch that passes my review criteria.

Also, identifying bad vibe-coding patches is usually easy: their descriptions look like written by a former literature professor that just took his first job as junior programmer: they come with useless comments, random variable renames, unneeded code rewrites, etc.

Btw, if one wants them to be tagged as AI-produced, perhaps we can teach Sashiko to identify it and add a big warning on its review. This will probably be more reliable than asking people to add a LLM tag.

What I think we should do, instead, is to let it clear that vibe coding is forbidden - e.g. just asking AI to do something and sending it as-is (*).
So, I would add something like this at Documentation/process/coding-assistants.rst:

"All patches shall be written or reviewed and modified by a human.
You should be able to understand exactly what the patch is doing, why it is taking such approach and how. Pure non-trivial machine-generated patches will be rejected."

(*) There are of course exceptions for it: typo fix patches for instance
can be pure machine-generated. Even before AI, we have patches like that where people used some spell checker to fix typos. Again, they need human review before submission, as sometimes, spell checker fails when, for instance, it tries to replace an acronym with resembles a word.

Signature · 2 lines
Thanks,
Mauro
#32 lore raw
In reply to Greg KH
Quoted 30 lines · Greg KH
Quoted 29 lines · Laurent Pinchart
Quoted 27 lines · Jeff Layton
Quoted 12 lines · Greg KH
Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Then let's move to get people to follow it.

Quoted 1 line

2/ It basically serves as free advertising for proprietary LLM companies.

Quoted 8 lines

Who cares, make up a name, all I want is the "signal" that someone is using a LLM so that I can review it as-such. And if I think someone is not reporting that, I can ask for them to properly attribute it and if they lie, well, that's on them.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

We want to know if a LLM is being used.

But why? What do you intend to do with this information?

Do you mean to use it as an indicator that the patch should receive "extra" review (or maybe that it should be ignored)? Do you mean to use it to generate some sort of statistics at a later time?

Quoted 9 lines
Quoted 3 lines

I use the information to decide how to review the patch, and what level of priority to give it. For that usage I don't need a tag, but I need the information in some human-readable form at patch submission time.

Same here. I don't care about stats, I care about "how do I review this patch" and this gives me that signal that I need if faced with a llm-helped patch.

Do we need a tag for this though?

This seems like the kind of information that we would always require in the cover letter of a series (or the little place in an individual patch for comments that don't get merged). That would also allow you to convey a lot more nuance about how it was used.

ISTM asking people to disclose LLM usage in a cover letter would give everyone what they want: Information about whether and possibly how an LLM was used, and it also wouldn't clutter up the changelogs with these tags.

Signature · 1 line
Jeff Layton <[email protected]>
#33 lore raw
Quoted 48 lines · Jeff Layton
Quoted 30 lines · Greg KH
Quoted 29 lines · Laurent Pinchart
Quoted 27 lines · Jeff Layton
Quoted 12 lines · Greg KH
Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Then let's move to get people to follow it.

Quoted 1 line

2/ It basically serves as free advertising for proprietary LLM companies.

Quoted 8 lines

Who cares, make up a name, all I want is the "signal" that someone is using a LLM so that I can review it as-such. And if I think someone is not reporting that, I can ask for them to properly attribute it and if they lie, well, that's on them.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

We want to know if a LLM is being used.

But why? What do you intend to do with this information?

Do you mean to use it as an indicator that the patch should receive "extra" review (or maybe that it should be ignored)? Do you mean to use it to generate some sort of statistics at a later time?

Quoted 9 lines
Quoted 3 lines

I use the information to decide how to review the patch, and what level of priority to give it. For that usage I don't need a tag, but I need the information in some human-readable form at patch submission time.

Same here. I don't care about stats, I care about "how do I review this patch" and this gives me that signal that I need if faced with a llm-helped patch.

Do we need a tag for this though?

This seems like the kind of information that we would always require in the cover letter of a series (or the little place in an individual patch for comments that don't get merged). That would also allow you to convey a lot more nuance about how it was used.

Quoted 4 lines

ISTM asking people to disclose LLM usage in a cover letter would give everyone what they want: Information about whether and possibly how an LLM was used, and it also wouldn't clutter up the changelogs with these tags.

It's much much clearer and easier to just have a standardised tag for that.

You can see that (and grep for that) immediately, vague paragraphs not so much.

Quoted 2 lines

--
Jeff Layton <[email protected]>

Thanks, Lroenzo

#34 lore raw
Quoted 49 lines · Lorenzo Stoakes
Quoted 48 lines · Jeff Layton
Quoted 30 lines · Greg KH
Quoted 29 lines · Laurent Pinchart
Quoted 27 lines · Jeff Layton
Quoted 12 lines · Greg KH
Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Then let's move to get people to follow it.

Quoted 1 line

2/ It basically serves as free advertising for proprietary LLM companies.

Quoted 8 lines

Who cares, make up a name, all I want is the "signal" that someone is using a LLM so that I can review it as-such. And if I think someone is not reporting that, I can ask for them to properly attribute it and if they lie, well, that's on them.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

We want to know if a LLM is being used.

But why? What do you intend to do with this information?

Do you mean to use it as an indicator that the patch should receive "extra" review (or maybe that it should be ignored)? Do you mean to use it to generate some sort of statistics at a later time?

Quoted 9 lines
Quoted 3 lines

I use the information to decide how to review the patch, and what level of priority to give it. For that usage I don't need a tag, but I need the information in some human-readable form at patch submission time.

Same here. I don't care about stats, I care about "how do I review this patch" and this gives me that signal that I need if faced with a llm-helped patch.

Do we need a tag for this though?

This seems like the kind of information that we would always require in the cover letter of a series (or the little place in an individual patch for comments that don't get merged). That would also allow you to convey a lot more nuance about how it was used.

Quoted 9 lines
Quoted 4 lines

ISTM asking people to disclose LLM usage in a cover letter would give everyone what they want: Information about whether and possibly how an LLM was used, and it also wouldn't clutter up the changelogs with these tags.

It's much much clearer and easier to just have a standardised tag for that.

You can see that (and grep for that) immediately, vague paragraphs not so much.

At the risk of being pedantic on a point where I think the document is kind of lacking:

What level of assistance crosses the bar for an "Assisted-by: LLM" tag?

Some sample levels of assistance to illustrate the point:

1. I used an llm to one-shot vibe-code a patch
2. I used an llm to write a patch but carefully reviewed every line
3. I used an llm to explore the design space for a patch but wrote it manually
4. I used an llm to debug or reproduce a kernel issue but then wrote the fix manually after fully understanding the defect
5. I used an llm to review a patch I wrote
6. I used an llm to research some chunk of code while writing a patch
7. I used Google while writing a patch and learned something valuable from the AI overview at the top

I personally would 100% use the tag for 1 or 2, and have already done so. I have not been doing it for 3-5, as I think that will basically make every patch llm-assisted to the point of the distinction being meaningless. If we should be doing it for 3-5 (or some subset thereof) then my mistake and I will certainly start doing so. I would hope most people agree 6-7 and similar need no tag.

Similar questions abound if you use an llm to help with writing the English text in the patch or emails.

I have a feeling that this ambiguity is part of the reason we aren't all agreeing on the value of the tag?

Thanks,
Boris

Quoted 4 lines
Quoted 2 lines

--
Jeff Layton <[email protected]>

Thanks, Lroenzo

#35 lore raw
Quoted 78 lines · On Thu, 2 Jul 2026 14:17:40 -0700 Boris Burkov <[email protected]> wrote:
Quoted 49 lines · Lorenzo Stoakes
Quoted 48 lines · Jeff Layton
Quoted 30 lines · Greg KH
Quoted 29 lines · Laurent Pinchart
Quoted 27 lines · Jeff Layton
Quoted 12 lines · Greg KH
Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Then let's move to get people to follow it.

Quoted 1 line

2/ It basically serves as free advertising for proprietary LLM companies.

Quoted 8 lines

Who cares, make up a name, all I want is the "signal" that someone is using a LLM so that I can review it as-such. And if I think someone is not reporting that, I can ask for them to properly attribute it and if they lie, well, that's on them.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

We want to know if a LLM is being used.

But why? What do you intend to do with this information?

Do you mean to use it as an indicator that the patch should receive "extra" review (or maybe that it should be ignored)? Do you mean to use it to generate some sort of statistics at a later time?

Quoted 9 lines
Quoted 3 lines

I use the information to decide how to review the patch, and what level of priority to give it. For that usage I don't need a tag, but I need the information in some human-readable form at patch submission time.

Same here. I don't care about stats, I care about "how do I review this patch" and this gives me that signal that I need if faced with a llm-helped patch.

Do we need a tag for this though?

This seems like the kind of information that we would always require in the cover letter of a series (or the little place in an individual patch for comments that don't get merged). That would also allow you to convey a lot more nuance about how it was used.

Quoted 9 lines
Quoted 4 lines

ISTM asking people to disclose LLM usage in a cover letter would give everyone what they want: Information about whether and possibly how an LLM was used, and it also wouldn't clutter up the changelogs with these tags.

It's much much clearer and easier to just have a standardised tag for that.

You can see that (and grep for that) immediately, vague paragraphs not so much.

At the risk of being pedantic on a point where I think the document is kind of lacking:

What level of assistance crosses the bar for an "Assisted-by: LLM" tag?

Some sample levels of assistance to illustrate the point:

1. I used an llm to one-shot vibe-code a patch
2. I used an llm to write a patch but carefully reviewed every line
3. I used an llm to explore the design space for a patch but wrote it manually
4. I used an llm to debug or reproduce a kernel issue but then wrote the fix manually after fully understanding the defect
5. I used an llm to review a patch I wrote
6. I used an llm to research some chunk of code while writing a patch
7. I used Google while writing a patch and learned something valuable from the AI overview at the top

Quoted 6 lines

I personally would 100% use the tag for 1 or 2, and have already done so. I have not been doing it for 3-5, as I think that will basically make every patch llm-assisted to the point of the distinction being meaningless. If we should be doing it for 3-5 (or some subset thereof) then my mistake and I will certainly start doing so. I would hope most people agree 6-7 and similar need no tag.

I think 3-7 don't need the tag.

Quoted 3 lines

Similar questions abound if you use an llm to help with writing the English text in the patch or emails.

I think this doesn't need the tag, too. Like patches, the user should completely reviewed and understand the text, though. Without the complete review, it should not be submmitted at all, regardless of the tag.

Quoted 3 lines

I have a feeling that this ambiguity is part of the reason we aren't all agreeing on the value of the tag?

I'm not very sure... I don't really feel the question is that difficult and ambiguous to answer. I may be biased.

Thanks,
SJ

[...]

#36 lore raw
In reply to Boris Burkov
Quoted 45 lines · Boris Burkov
Quoted 9 lines · Lorenzo Stoakes
Quoted 8 lines · Jeff Layton

Do we need a tag for this though?

This seems like the kind of information that we would always require in the cover letter of a series (or the little place in an individual patch for comments that don't get merged). That would also allow you to convey a lot more nuance about how it was used.

Quoted 9 lines
Quoted 4 lines

ISTM asking people to disclose LLM usage in a cover letter would give everyone what they want: Information about whether and possibly how an LLM was used, and it also wouldn't clutter up the changelogs with these tags.

It's much much clearer and easier to just have a standardised tag for that.

You can see that (and grep for that) immediately, vague paragraphs not so much.

At the risk of being pedantic on a point where I think the document is kind of lacking:

What level of assistance crosses the bar for an "Assisted-by: LLM" tag?

Some sample levels of assistance to illustrate the point:

1. I used an llm to one-shot vibe-code a patch
2. I used an llm to write a patch but carefully reviewed every line
3. I used an llm to explore the design space for a patch but wrote it manually
4. I used an llm to debug or reproduce a kernel issue but then wrote the fix manually after fully understanding the defect
5. I used an llm to review a patch I wrote
6. I used an llm to research some chunk of code while writing a patch
7. I used Google while writing a patch and learned something valuable from the AI overview at the top

I personally would 100% use the tag for 1 or 2, and have already done so. I have not been doing it for 3-5, as I think that will basically make every patch llm-assisted to the point of the distinction being meaningless. If we should be doing it for 3-5 (or some subset thereof) then my mistake and I will certainly start doing so. I would hope most people agree 6-7 and similar need no tag.

Quoted 5 lines

Similar questions abound if you use an llm to help with writing the English text in the patch or emails.

I have a feeling that this ambiguity is part of the reason we aren't all agreeing on the value of the tag?

Yes, I raised something similar as reply to Christian's RFC [1], where I said that for me the information *how* it was used is much more important:

"

Assisted-by LLM # translate commit message
Assisted-by LLM # generate some test cases
Assisted-by LLM # cleanup logic
Assisted-by LLM # everything and I have no clue what any in here does

"

That tag is it stands is pretty useless, really.

I assume most people only really use it for something in-between 1 and 2, but *who knows*.

[1] https://lore.kernel.org/r/[email protected]

Signature · 3 lines
Cheers,

David
#37 lore raw
Quoted 39 lines · David Hildenbrand (Arm)
Quoted 38 lines · Boris Burkov
Quoted 9 lines · Lorenzo Stoakes
Quoted 8 lines · Jeff Layton

Do we need a tag for this though?

This seems like the kind of information that we would always require in the cover letter of a series (or the little place in an individual patch for comments that don't get merged). That would also allow you to convey a lot more nuance about how it was used.

Quoted 9 lines
Quoted 4 lines

ISTM asking people to disclose LLM usage in a cover letter would give everyone what they want: Information about whether and possibly how an LLM was used, and it also wouldn't clutter up the changelogs with these tags.

It's much much clearer and easier to just have a standardised tag for that.

You can see that (and grep for that) immediately, vague paragraphs not so much.

At the risk of being pedantic on a point where I think the document is kind of lacking:

What level of assistance crosses the bar for an "Assisted-by: LLM" tag?

Some sample levels of assistance to illustrate the point:

1. I used an llm to one-shot vibe-code a patch
2. I used an llm to write a patch but carefully reviewed every line
3. I used an llm to explore the design space for a patch but wrote it manually
4. I used an llm to debug or reproduce a kernel issue but then wrote the fix manually after fully understanding the defect
5. I used an llm to review a patch I wrote
6. I used an llm to research some chunk of code while writing a patch
7. I used Google while writing a patch and learned something valuable from the AI overview at the top

Quoted 6 lines
Quoted 6 lines

I personally would 100% use the tag for 1 or 2, and have already done so. I have not been doing it for 3-5, as I think that will basically make every patch llm-assisted to the point of the distinction being meaningless. If we should be doing it for 3-5 (or some subset thereof) then my mistake and I will certainly start doing so. I would hope most people agree 6-7 and similar need no tag.

I personally think 1-2 are the only relevant cases.

Quoted 10 lines
Quoted 6 lines

Similar questions abound if you use an llm to help with writing the English text in the patch or emails.

I have a feeling that this ambiguity is part of the reason we aren't all agreeing on the value of the tag?

Yes, I raised something similar as reply to Christian's RFC [1], where I said that for me the information *how* it was used is much more important:

Quoted 6 lines

"

Assisted-by LLM # translate commit message
Assisted-by LLM # generate some test cases
Assisted-by LLM # cleanup logic
Assisted-by LLM # everything and I have no clue what any in here does

"

Yup, and we don't need complicated rules for that just 'document what you used it for and give a sense of how much'.

It's fuzzy but useful.

Quoted 2 lines

That tag is it stands is pretty useless, really.

Not to go over it all again but I disagree, even as it stands, it allows us to engage in conversation about the LLM usage if admitted, and to point those who are misbehaving at the rules if not.

And it is a clear way to get the boolean 'is this person saying they used an LLM'.

But I agree with you it'd be MUCH more useful if we did the above.

I wonder if we could get consensus on adding a section to the doc saying that it'd be _useful_ to add a comment explaining _what_ you did, and explaining the concept with some examples?

I can't imagine anybody would disagree with that, and that would get us positive forward progress.

Then later we could debate the details further?

Quoted 10 lines

I assume most people only really use it for something in-between 1 and 2, but *who knows*.

[1] https://lore.kernel.org/r/[email protected]

--
Cheers,

David

Thanks, Lorenzo

#38 lore raw
Quoted 78 lines · Lorenzo Stoakes
Quoted 39 lines · David Hildenbrand (Arm)
Quoted 38 lines · Boris Burkov
Quoted 9 lines · Lorenzo Stoakes
Quoted 8 lines · Jeff Layton

Do we need a tag for this though?

This seems like the kind of information that we would always require in the cover letter of a series (or the little place in an individual patch for comments that don't get merged). That would also allow you to convey a lot more nuance about how it was used.

Quoted 9 lines
Quoted 4 lines

ISTM asking people to disclose LLM usage in a cover letter would give everyone what they want: Information about whether and possibly how an LLM was used, and it also wouldn't clutter up the changelogs with these tags.

It's much much clearer and easier to just have a standardised tag for that.

You can see that (and grep for that) immediately, vague paragraphs not so much.

At the risk of being pedantic on a point where I think the document is kind of lacking:

What level of assistance crosses the bar for an "Assisted-by: LLM" tag?

Some sample levels of assistance to illustrate the point:

1. I used an llm to one-shot vibe-code a patch
2. I used an llm to write a patch but carefully reviewed every line
3. I used an llm to explore the design space for a patch but wrote it manually
4. I used an llm to debug or reproduce a kernel issue but then wrote the fix manually after fully understanding the defect
5. I used an llm to review a patch I wrote
6. I used an llm to research some chunk of code while writing a patch
7. I used Google while writing a patch and learned something valuable from the AI overview at the top

Quoted 6 lines
Quoted 6 lines

I personally would 100% use the tag for 1 or 2, and have already done so. I have not been doing it for 3-5, as I think that will basically make every patch llm-assisted to the point of the distinction being meaningless. If we should be doing it for 3-5 (or some subset thereof) then my mistake and I will certainly start doing so. I would hope most people agree 6-7 and similar need no tag.

I personally think 1-2 are the only relevant cases.

Quoted 10 lines
Quoted 6 lines

Similar questions abound if you use an llm to help with writing the English text in the patch or emails.

I have a feeling that this ambiguity is part of the reason we aren't all agreeing on the value of the tag?

Yes, I raised something similar as reply to Christian's RFC [1], where I said that for me the information *how* it was used is much more important:

Quoted 6 lines

"

Assisted-by LLM # translate commit message
Assisted-by LLM # generate some test cases
Assisted-by LLM # cleanup logic
Assisted-by LLM # everything and I have no clue what any in here does

"

Yup, and we don't need complicated rules for that just 'document what you used it for and give a sense of how much'.

It's fuzzy but useful.

Quoted 2 lines

That tag is it stands is pretty useless, really.

Not to go over it all again but I disagree, even as it stands, it allows us to engage in conversation about the LLM usage if admitted, and to point those who are misbehaving at the rules if not.

Quoted 8 lines

And it is a clear way to get the boolean 'is this person saying they used an LLM'.

But I agree with you it'd be MUCH more useful if we did the above.

I wonder if we could get consensus on adding a section to the doc saying that it'd be _useful_ to add a comment explaining _what_ you did, and explaining the concept with some examples?

I'd support a patch that replaces

Assisted-by Claude:claude-3-opus coccinelle sparse

with

Assisted-by LLM # generate some test cases

and rewrites the Attribution section of Documentation/process/coding-assistants.rst accordingly.

I think most people in this mail thread have expressed that how generative AI was used is the most important information, and several people (including myself) have expressed a desire to stop the free advertising. Unless I missed something, I don't think anyone has expressed an interest in keeping the agent name and model.

Quoted 9 lines

I can't imagine anybody would disagree with that, and that would get us positive forward progress.

Then later we could debate the details further?

Quoted 4 lines

I assume most people only really use it for something in-between 1 and 2, but *who knows*.

[1] https://lore.kernel.org/r/[email protected]

Signature · 3 lines
Regards,

Laurent Pinchart
#39 lore raw
Quoted 45 lines · Laurent Pinchart
Quoted 27 lines · Lorenzo Stoakes

On Fri, Jul 03, 2026 at 09:05:58AM +0200, David Hildenbrand (Arm) wrote:

I personally think 1-2 are the only relevant cases.

Quoted 10 lines

Yes, I raised something similar as reply to Christian's RFC [1], where I said that for me the information *how* it was used is much more important:

"

Assisted-by LLM # translate commit message
Assisted-by LLM # generate some test cases
Assisted-by LLM # cleanup logic
Assisted-by LLM # everything and I have no clue what any in here does

"

Yup, and we don't need complicated rules for that just 'document what you used it for and give a sense of how much'.

It's fuzzy but useful.

Quoted 2 lines

That tag is it stands is pretty useless, really.

Not to go over it all again but I disagree, even as it stands, it allows us to engage in conversation about the LLM usage if admitted, and to point those who are misbehaving at the rules if not.

Quoted 8 lines

And it is a clear way to get the boolean 'is this person saying they used an LLM'.

But I agree with you it'd be MUCH more useful if we did the above.

I wonder if we could get consensus on adding a section to the doc saying that it'd be _useful_ to add a comment explaining _what_ you did, and explaining the concept with some examples?

I'd support a patch that replaces

Assisted-by Claude:claude-3-opus coccinelle sparse

with

Assisted-by LLM # generate some test cases
Quoted 8 lines

and rewrites the Attribution section of Documentation/process/coding-assistants.rst accordingly.

I think most people in this mail thread have expressed that how generative AI was used is the most important information, and several people (including myself) have expressed a desire to stop the free advertising. Unless I missed something, I don't think anyone has expressed an interest in keeping the agent name and model.

Yes, that's something I would enjoy seeing. Who reading along has a problem with that? (assuming that the tag must stay)

Signature · 3 lines
Cheers,

David
#40 lore raw
Quoted 49 lines · David Hildenbrand (Arm)
Quoted 48 lines · Laurent Pinchart
Quoted 27 lines · Lorenzo Stoakes

On Fri, Jul 03, 2026 at 09:05:58AM +0200, David Hildenbrand (Arm) wrote:

I personally think 1-2 are the only relevant cases.

Quoted 10 lines

Yes, I raised something similar as reply to Christian's RFC [1], where I said that for me the information *how* it was used is much more important:

"

Assisted-by LLM # translate commit message
Assisted-by LLM # generate some test cases
Assisted-by LLM # cleanup logic
Assisted-by LLM # everything and I have no clue what any in here does

"

Yup, and we don't need complicated rules for that just 'document what you used it for and give a sense of how much'.

It's fuzzy but useful.

Quoted 2 lines

That tag is it stands is pretty useless, really.

Not to go over it all again but I disagree, even as it stands, it allows us to engage in conversation about the LLM usage if admitted, and to point those who are misbehaving at the rules if not.

Quoted 8 lines

And it is a clear way to get the boolean 'is this person saying they used an LLM'.

But I agree with you it'd be MUCH more useful if we did the above.

I wonder if we could get consensus on adding a section to the doc saying that it'd be _useful_ to add a comment explaining _what_ you did, and explaining the concept with some examples?

I'd support a patch that replaces

Assisted-by Claude:claude-3-opus coccinelle sparse

with

Assisted-by LLM # generate some test cases

and rewrites the Attribution section of Documentation/process/coding-assistants.rst accordingly.

Quoted 8 lines
Quoted 5 lines

I think most people in this mail thread have expressed that how generative AI was used is the most important information, and several people (including myself) have expressed a desire to stop the free advertising. Unless I missed something, I don't think anyone has expressed an interest in keeping the agent name and model.

Yes, that's something I would enjoy seeing. Who reading along has a problem with that? (assuming that the tag must stay)

I think the silence suggests nobody :)

What we could do is send a patch to simply add a section on adding a comment explaining _what_ used an LLM and strongly suggesting it would be helpful to do so.

That should hopefully be uncontroversial, then later (or alongside that?) we could submit an RFC for making the language move to 'you MUST do this', as that will probably be the more controversial bit.

I don't see any reason why we can't just send the softer version right away.

Quoted 5 lines

--
Cheers,

David

Cheers, Lorenzo

#41 lore raw
Quoted 25 lines · Lorenzo Stoakes
Quoted 13 lines · David Hildenbrand (Arm)
Quoted 12 lines · Laurent Pinchart

I'd support a patch that replaces

Assisted-by Claude:claude-3-opus coccinelle sparse

with

Assisted-by LLM # generate some test cases

and rewrites the Attribution section of Documentation/process/coding-assistants.rst accordingly.

Quoted 8 lines
Quoted 5 lines

I think most people in this mail thread have expressed that how generative AI was used is the most important information, and several people (including myself) have expressed a desire to stop the free advertising. Unless I missed something, I don't think anyone has expressed an interest in keeping the agent name and model.

Yes, that's something I would enjoy seeing. Who reading along has a problem with that? (assuming that the tag must stay)

I think the silence suggests nobody :)

Quoted 7 lines

What we could do is send a patch to simply add a section on adding a comment explaining _what_ used an LLM and strongly suggesting it would be helpful to do so.

That should hopefully be uncontroversial, then later (or alongside that?) we could submit an RFC for making the language move to 'you MUST do this', as that will probably be the more controversial bit.

Maybe we should just spell out about which scenario we care about where we really expect the disclosure.

Sure, it might be more controversial and we could have that in a separate patch.

I was thinking of the following end result:

Documentation/process/coding-assistants.rst +25 -8
index 899f4459c52d2..4c2ab4dfc6da7 100644
@@ -40,20 +40,37 @@ Attribution
 ===========
 
 When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
+helps track the evolving role of AI in the development process. Further,
+for reviewers and maintainers it is also crucially important to know how
+AI tools were used.
 
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
+Contributions that used AI to generate significant portions of code,
+comments, or patch descriptions must include an Assisted-by tag in the
+following format::
 
-Where (preferred):
+  Assisted-by: LLM # brief description of usage
+
+Or alternatively::
+
+  Assisted-by: AGENT_NAME:MODEL_VERSION # brief description of usage
+
+Where::
 
 * ``AGENT_NAME`` is the name of the AI tool or framework
 * ``MODEL_VERSION`` is the specific model version used
-* ``[TOOL1] [TOOL2]`` are optional specialized analysis tools used
-  (e.g., coccinelle, sparse, smatch, clang-tidy)
+
+If other tools were used, they should be specified through a dedicated
+Assisted-by tag in the following format::
+
+  Assisted-by: [TOOL1] [TOOL2]
+
+Where ``[TOOL1] [TOOL2]`` are specialized analysis tools used
+(e.g., coccinelle, sparse, smatch, clang-tidy)
 
 Basic development tools (git, gcc, make, editors) should not be listed.
 
-Example::
+Examples::
 
-  Assisted-by: Claude:claude-3-opus coccinelle sparse
+  Assisted-by: LLM # translate patch description
+  Assisted-by: Claude:claude-3-opus # generate most of the code
+  Assisted-by: coccinelle sparse

But I am not really sure why we should specify other tools here, really. People usually do that as part of the patch description, including sharing coccinelle scripts etc.

Because looking at the history:

$ git log | grep "Assisted-by" | grep -E "(coccinelle|sparse|smatch|clang)"
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
Signature · 3 lines
Cheers,

David
#42 lore raw
Quoted 81 lines · David Hildenbrand (Arm)
Quoted 25 lines · Lorenzo Stoakes
Quoted 13 lines · David Hildenbrand (Arm)
Quoted 12 lines · Laurent Pinchart

I'd support a patch that replaces

Assisted-by Claude:claude-3-opus coccinelle sparse

with

Assisted-by LLM # generate some test cases

and rewrites the Attribution section of Documentation/process/coding-assistants.rst accordingly.

Quoted 8 lines
Quoted 5 lines

I think most people in this mail thread have expressed that how generative AI was used is the most important information, and several people (including myself) have expressed a desire to stop the free advertising. Unless I missed something, I don't think anyone has expressed an interest in keeping the agent name and model.

Yes, that's something I would enjoy seeing. Who reading along has a problem with that? (assuming that the tag must stay)

I think the silence suggests nobody :)

Quoted 7 lines

What we could do is send a patch to simply add a section on adding a comment explaining _what_ used an LLM and strongly suggesting it would be helpful to do so.

That should hopefully be uncontroversial, then later (or alongside that?) we could submit an RFC for making the language move to 'you MUST do this', as that will probably be the more controversial bit.

Maybe we should just spell out about which scenario we care about where we really expect the disclosure.

Sure, it might be more controversial and we could have that in a separate patch.

I was thinking of the following end result:

Documentation/process/coding-assistants.rst +21 -6
index 899f4459c52d2..4c2ab4dfc6da7 100644
@@ -40,20 +40,37 @@ Attribution
 ===========
 
 When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
+helps track the evolving role of AI in the development process. Further,
+for reviewers and maintainers it is also crucially important to know how
+AI tools were used.
 
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
+Contributions that used AI to generate significant portions of code,
+comments, or patch descriptions must include an Assisted-by tag in the
+following format::
 
-Where (preferred):
+  Assisted-by: LLM # brief description of usage
+
+Or alternatively::
+
+  Assisted-by: AGENT_NAME:MODEL_VERSION # brief description of usage
+
+Where::
 
 * ``AGENT_NAME`` is the name of the AI tool or framework
 * ``MODEL_VERSION`` is the specific model version used
-* ``[TOOL1] [TOOL2]`` are optional specialized analysis tools used
-  (e.g., coccinelle, sparse, smatch, clang-tidy)
+
+If other tools were used, they should be specified through a dedicated
+Assisted-by tag in the following format::
+
+  Assisted-by: [TOOL1] [TOOL2]
+
+Where ``[TOOL1] [TOOL2]`` are specialized analysis tools used
+(e.g., coccinelle, sparse, smatch, clang-tidy)

Documentation/process/coding-assistants.rst Attribution · line 76

Quoted 10 lines
Documentation/process/coding-assistants.rst +4 -2
 
 Basic development tools (git, gcc, make, editors) should not be listed.
 
-Example::
+Examples::
 
-  Assisted-by: Claude:claude-3-opus coccinelle sparse
+  Assisted-by: LLM # translate patch description
+  Assisted-by: Claude:claude-3-opus # generate most of the code
+  Assisted-by: coccinelle sparse

LGTM! Do you want to send that then? People can comment on the actual path then (probably worth cc'ing everybody here on that also).

Quoted 8 lines

But I am not really sure why we should specify other tools here, really. People usually do that as part of the patch description, including sharing coccinelle scripts etc.

Because looking at the history:

(for succinctness :P)

$ git log | grep "Assisted-by" | grep -Ei "(coccinelle|sparse|smatch|clang)" | wc -l
30

Yeah, doesn't seem much used, but I guess for the non-controversial version worth keeping that stuff around in the doc.

Quoted 4 lines

--
Cheers,

David

Cheers, Lorenzo

#43 lore raw
Quoted 66 lines · Lorenzo Stoakes
Quoted 65 lines · David Hildenbrand (Arm)
Quoted 10 lines · Lorenzo Stoakes

I think the silence suggests nobody :)

What we could do is send a patch to simply add a section on adding a comment explaining _what_ used an LLM and strongly suggesting it would be helpful to do so.

That should hopefully be uncontroversial, then later (or alongside that?) we could submit an RFC for making the language move to 'you MUST do this', as that will probably be the more controversial bit.

Maybe we should just spell out about which scenario we care about where we really expect the disclosure.

Sure, it might be more controversial and we could have that in a separate patch.

I was thinking of the following end result:

Documentation/process/coding-assistants.rst +22 -7
index 899f4459c52d2..4c2ab4dfc6da7 100644
@@ -40,20 +40,37 @@ Attribution
 ===========
 
 When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
+helps track the evolving role of AI in the development process. Further,
+for reviewers and maintainers it is also crucially important to know how
+AI tools were used.
 
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
+Contributions that used AI to generate significant portions of code,
+comments, or patch descriptions must include an Assisted-by tag in the
+following format::
 
-Where (preferred):
+  Assisted-by: LLM # brief description of usage
+
+Or alternatively::
+
+  Assisted-by: AGENT_NAME:MODEL_VERSION # brief description of usage
+
+Where::
 
 * ``AGENT_NAME`` is the name of the AI tool or framework
 * ``MODEL_VERSION`` is the specific model version used
-* ``[TOOL1] [TOOL2]`` are optional specialized analysis tools used
-  (e.g., coccinelle, sparse, smatch, clang-tidy)
+
+If other tools were used, they should be specified through a dedicated
+Assisted-by tag in the following format::
+
+  Assisted-by: [TOOL1] [TOOL2]
+
+Where ``[TOOL1] [TOOL2]`` are specialized analysis tools used
+(e.g., coccinelle, sparse, smatch, clang-tidy)
 
 Basic development tools (git, gcc, make, editors) should not be listed.
 
-Example::
+Examples::
 
Quoted 7 lines

Documentation/process/coding-assistants.rst Attribution · line 76

Quoted 4 lines
Documentation/process/coding-assistants.rst +3 -1
-  Assisted-by: Claude:claude-3-opus coccinelle sparse
+  Assisted-by: LLM # translate patch description
+  Assisted-by: Claude:claude-3-opus # generate most of the code
+  Assisted-by: coccinelle sparse

LGTM! Do you want to send that then? People can comment on the actual path then (probably worth cc'ing everybody here on that also).

Let's first gather some more thoughts on the rough direction before spinning of yet another discussion. :)

Signature · 3 lines
Cheers,

David
#44 lore raw
Quoted 6 lines · David Hildenbrand (Arm)
Quoted 2 lines · Lorenzo Stoakes

LGTM! Do you want to send that then? People can comment on the actual path then (probably worth cc'ing everybody here on that also).

Let's first gather some more thoughts on the rough direction before spinning of yet another discussion. :)

I'm suggesting sending a simple, uncontroversial, change to get movement rather than continue this never-ending talking shop :)

Anyway I'll leave it up to you!

Thanks, Lorenzo

#45 lore raw
Quoted 10 lines · Lorenzo Stoakes
Quoted 6 lines · David Hildenbrand (Arm)
Quoted 2 lines · Lorenzo Stoakes

LGTM! Do you want to send that then? People can comment on the actual path then (probably worth cc'ing everybody here on that also).

Let's first gather some more thoughts on the rough direction before spinning of yet another discussion. :)

I'm suggesting sending a simple, uncontroversial, change to get movement rather than continue this never-ending talking shop :)

Oh but I love shop talk! :P

My opinions (having been on vacation for most of this thread) are roughly:

1. For patches generated by deterministic tools (e.g. sed/cocinelle), please include the source code so that anyone reading the patch can check the reproducibility of that patch. But that can be free-form in the commit message:

https://lore.kernel.org/linux-xfs/157343508488.1945685.9867882880040545380.stgit@magnolia/

2. For nondeterministic machine assistance, I like the idea of asking patch submitters to note which part(s) of the patch had machine assistance applied. I don't care all that much about the technology used (e.g. LLM, or Eliza, or whatever). Something like:

Assisted-by LLM # commit message
Assisted-by LLM # finding bugs in the original commit

I don't care to give free advertising to any specific LLM-pusher, nor do I care to give the whole *industry* any free publicity. They can spend their own leveraged money on advertising.

Mecha-assistance-done-by: finding bugs in the original commit

or maybe just

MAD: finding bugs in the original commit

:P

3. I grade *all* the trailers that submitters attach to patches! And how well they engage me on my weird followup questions! Both help me to construct a conscientiousness vibe, which is how I decide how much effort to put into making a response. Do you allege that your patch fixes a bug but fail to cc stable? Do you post obviously LLM generated content but leave out an Assisted-by tag? Do you habitually drop off the list for long periods of time? Decline to run /any/ QA on your patches at all? Not gonna waste my time.

OTOH, I also look for *good* behavior: cc'ing stable, responding to questions with a day or two, kvetching on fstests@ about the horridness of bash scripts? I find that endearing. ;)

--D

Quoted 4 lines

Anyway I'll leave it up to you!

Thanks, Lorenzo

#46 lore raw
Quoted 53 lines · Darrick J. Wong
Quoted 10 lines · Lorenzo Stoakes
Quoted 6 lines · David Hildenbrand (Arm)
Quoted 2 lines · Lorenzo Stoakes

LGTM! Do you want to send that then? People can comment on the actual path then (probably worth cc'ing everybody here on that also).

Let's first gather some more thoughts on the rough direction before spinning of yet another discussion. :)

I'm suggesting sending a simple, uncontroversial, change to get movement rather than continue this never-ending talking shop :)

Oh but I love shop talk! :P

My opinions (having been on vacation for most of this thread) are roughly:

1. For patches generated by deterministic tools (e.g. sed/cocinelle), please include the source code so that anyone reading the patch can check the reproducibility of that patch. But that can be free-form in the commit message:

https://lore.kernel.org/linux-xfs/157343508488.1945685.9867882880040545380.stgit@magnolia/

2. For nondeterministic machine assistance, I like the idea of asking patch submitters to note which part(s) of the patch had machine assistance applied. I don't care all that much about the technology used (e.g. LLM, or Eliza, or whatever). Something like:

Assisted-by LLM # commit message
Assisted-by LLM # finding bugs in the original commit

I don't care to give free advertising to any specific LLM-pusher, nor do I care to give the whole *industry* any free publicity. They can spend their own leveraged money on advertising.

Mecha-assistance-done-by: finding bugs in the original commit

or maybe just

MAD: finding bugs in the original commit

:P

3. I grade *all* the trailers that submitters attach to patches! And how well they engage me on my weird followup questions! Both help me to construct a conscientiousness vibe, which is how I decide how much effort to put into making a response. Do you allege that your patch fixes a bug but fail to cc stable? Do you post obviously LLM generated content but leave out an Assisted-by tag? Do you habitually drop off the list for long periods of time? Decline to run /any/ QA on your patches at all? Not gonna waste my time.

Quoted 10 lines

OTOH, I also look for *good* behavior: cc'ing stable, responding to questions with a day or two, kvetching on fstests@ about the horridness of bash scripts? I find that endearing. ;)

--D

Quoted 4 lines

Anyway I'll leave it up to you!

Thanks, Lorenzo

I love all of this, no notes :)

Cheers, Lorenzo

#47 lore raw
In reply to David Hildenbrand (Arm)

Hi David,

Thanks for your patch!

Quoted 28 lines · David Hildenbrand (Arm)
Documentation/process/coding-assistants.rst +13 -4
@@ -40,20 +40,37 @@ Attribution
 ===========
 
 When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
+helps track the evolving role of AI in the development process. Further,
+for reviewers and maintainers it is also crucially important to know how
+AI tools were used.
 
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
+Contributions that used AI to generate significant portions of code,
+comments, or patch descriptions must include an Assisted-by tag in the
+following format::
 
-Where (preferred):
+  Assisted-by: LLM # brief description of usage
+
+Or alternatively::
+
+  Assisted-by: AGENT_NAME:MODEL_VERSION # brief description of usage
+
+Where::
 
 * ``AGENT_NAME`` is the name of the AI tool or framework
 * ``MODEL_VERSION`` is the specific model version used

Documentation/process/coding-assistants.rst Attribution · line 68

Quoted 10 lines
Documentation/process/coding-assistants.rst +8 -2
-* ``[TOOL1] [TOOL2]`` are optional specialized analysis tools used
-  (e.g., coccinelle, sparse, smatch, clang-tidy)
+
+If other tools were used, they should be specified through a dedicated
+Assisted-by tag in the following format::
+
+  Assisted-by: [TOOL1] [TOOL2]
+
+Where ``[TOOL1] [TOOL2]`` are specialized analysis tools used
+(e.g., coccinelle, sparse, smatch, clang-tidy)

Unlike LLMs above, all of these are deterministic...

Documentation/process/coding-assistants.rst Attribution · line 70

Quoted 2 lines
Documentation/process/coding-assistants.rst +0 -0
 Basic development tools (git, gcc, make, editors) should not be listed.

... just like these.
So why treat them different?

Gr{oetje,eeting}s,

Geert

Signature · 5 lines
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds
#48 lore raw
Quoted 47 lines · Geert Uytterhoeven

Hi David,

Thanks for your patch!

Quoted 28 lines · David Hildenbrand (Arm)
Documentation/process/coding-assistants.rst +13 -4
@@ -40,20 +40,37 @@ Attribution
 ===========
 
 When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
+helps track the evolving role of AI in the development process. Further,
+for reviewers and maintainers it is also crucially important to know how
+AI tools were used.
 
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
+Contributions that used AI to generate significant portions of code,
+comments, or patch descriptions must include an Assisted-by tag in the
+following format::
 
-Where (preferred):
+  Assisted-by: LLM # brief description of usage
+
+Or alternatively::
+
+  Assisted-by: AGENT_NAME:MODEL_VERSION # brief description of usage
+
+Where::
 
 * ``AGENT_NAME`` is the name of the AI tool or framework
 * ``MODEL_VERSION`` is the specific model version used

Documentation/process/coding-assistants.rst Attribution · line 68

Quoted 10 lines
Documentation/process/coding-assistants.rst +8 -2
-* ``[TOOL1] [TOOL2]`` are optional specialized analysis tools used
-  (e.g., coccinelle, sparse, smatch, clang-tidy)
+
+If other tools were used, they should be specified through a dedicated
+Assisted-by tag in the following format::
+
+  Assisted-by: [TOOL1] [TOOL2]
+
+Where ``[TOOL1] [TOOL2]`` are specialized analysis tools used
+(e.g., coccinelle, sparse, smatch, clang-tidy)

Unlike LLMs above, all of these are deterministic...

Quoted 1 line
Quoted 4 lines
Quoted 1 line

Basic development tools (git, gcc, make, editors) should not be listed.

... just like these.
So why treat them different?

I didn't write that original paragraph, so I really can't tell ... I mean, it is good practice to document usage of coccinelle at least, or when a problem was found through sparse.

Signature · 3 lines
Cheers,

David
#49 lore raw
In reply to David Hildenbrand (Arm)
Quoted 56 lines · David Hildenbrand (Arm)
Quoted 25 lines · Lorenzo Stoakes
Quoted 13 lines · David Hildenbrand (Arm)
Quoted 12 lines · Laurent Pinchart

I'd support a patch that replaces

Assisted-by Claude:claude-3-opus coccinelle sparse

with

Assisted-by LLM # generate some test cases

and rewrites the Attribution section of Documentation/process/coding-assistants.rst accordingly.

Quoted 8 lines
Quoted 5 lines

I think most people in this mail thread have expressed that how generative AI was used is the most important information, and several people (including myself) have expressed a desire to stop the free advertising. Unless I missed something, I don't think anyone has expressed an interest in keeping the agent name and model.

Yes, that's something I would enjoy seeing. Who reading along has a problem with that? (assuming that the tag must stay)

I think the silence suggests nobody :)

Quoted 7 lines

What we could do is send a patch to simply add a section on adding a comment explaining _what_ used an LLM and strongly suggesting it would be helpful to do so.

That should hopefully be uncontroversial, then later (or alongside that?) we could submit an RFC for making the language move to 'you MUST do this', as that will probably be the more controversial bit.

Maybe we should just spell out about which scenario we care about where we really expect the disclosure.

Sure, it might be more controversial and we could have that in a separate patch.

I was thinking of the following end result:

Documentation/process/coding-assistants.rst +3 -3
index 899f4459c52d2..4c2ab4dfc6da7 100644
@@ -40,20 +40,37 @@ Attribution
 ===========
 
 When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
+helps track the evolving role of AI in the development process. Further,
+for reviewers and maintainers it is also crucially important to know how
+AI tools were used.
 
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]

Documentation/process/coding-assistants.rst Attribution · line 55

Quoted 10 lines
Documentation/process/coding-assistants.rst +8 -1
+Contributions that used AI to generate significant portions of code,
+comments, or patch descriptions must include an Assisted-by tag in the
+following format::
 
-Where (preferred):
+  Assisted-by: LLM # brief description of usage
+
+Or alternatively::
+
+  Assisted-by: AGENT_NAME:MODEL_VERSION # brief description of usage

I would like to drop this one. As I understand, several people objected to including the product name in the git history as it is just free adverising, and nobody objected to dropping it.

Quoted 27 lines
Documentation/process/coding-assistants.rst +14 -4
+
+Where::
 
 * ``AGENT_NAME`` is the name of the AI tool or framework
 * ``MODEL_VERSION`` is the specific model version used
-* ``[TOOL1] [TOOL2]`` are optional specialized analysis tools used
-  (e.g., coccinelle, sparse, smatch, clang-tidy)
+
+If other tools were used, they should be specified through a dedicated
+Assisted-by tag in the following format::
+
+  Assisted-by: [TOOL1] [TOOL2]
+
+Where ``[TOOL1] [TOOL2]`` are specialized analysis tools used
+(e.g., coccinelle, sparse, smatch, clang-tidy)
 
 Basic development tools (git, gcc, make, editors) should not be listed.
 
-Example::
+Examples::
 
-  Assisted-by: Claude:claude-3-opus coccinelle sparse
+  Assisted-by: LLM # translate patch description
+  Assisted-by: Claude:claude-3-opus # generate most of the code
+  Assisted-by: coccinelle sparse
Quoted 3 lines

But I am not really sure why we should specify other tools here, really. People usually do that as part of the patch description, including sharing coccinelle scripts etc.

We've mentioned coccinelle usage in commit messages, with a copy of the semantic patch, for years and years and I'm not aware of anyone giving a technical reason why we need a tag instead. The only reason I see is to avoid pointing specifically at LLMs and normalize their usage as "just another tool". That's a political reason, not a technical one.

Quoted 32 lines

Because looking at the history:

$ git log | grep "Assisted-by" | grep -E "(coccinelle|sparse|smatch|clang)"
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
    Assisted-by: unnamed:deepseek-v3.2 coccinelle
Signature · 3 lines
Regards,

Laurent Pinchart
#50 lore raw
Quoted 39 lines · Laurent Pinchart
Quoted 38 lines · David Hildenbrand (Arm)
Quoted 10 lines · Lorenzo Stoakes

I think the silence suggests nobody :)

What we could do is send a patch to simply add a section on adding a comment explaining _what_ used an LLM and strongly suggesting it would be helpful to do so.

That should hopefully be uncontroversial, then later (or alongside that?) we could submit an RFC for making the language move to 'you MUST do this', as that will probably be the more controversial bit.

Maybe we should just spell out about which scenario we care about where we really expect the disclosure.

Sure, it might be more controversial and we could have that in a separate patch.

I was thinking of the following end result:

Documentation/process/coding-assistants.rst +6 -3
index 899f4459c52d2..4c2ab4dfc6da7 100644
@@ -40,20 +40,37 @@ Attribution
 ===========
 
 When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
+helps track the evolving role of AI in the development process. Further,
+for reviewers and maintainers it is also crucially important to know how
+AI tools were used.
 
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
+Contributions that used AI to generate significant portions of code,
+comments, or patch descriptions must include an Assisted-by tag in the
+following format::
 
Quoted 10 lines

Documentation/process/coding-assistants.rst Attribution · line 55

Quoted 6 lines
Documentation/process/coding-assistants.rst +5 -1
-Where (preferred):
+  Assisted-by: LLM # brief description of usage
+
+Or alternatively::
+
+  Assisted-by: AGENT_NAME:MODEL_VERSION # brief description of usage

I would like to drop this one. As I understand, several people objected to including the product name in the git history as it is just free adverising, and nobody objected to dropping it.

Yeah. I mean, we could go the following direction (dropping the models and encouraging also a description of usage for other tools), but not sure how much pushback we will get:

Documentation/process/coding-assistants.rst +17 -11
index 899f4459c52d2..a085a15d59990 100644
@@ -40,20 +40,26 @@ Attribution
 ===========
 
 When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
+helps track the evolving role of AI in the development process. Further,
+for reviewers and maintainers it is also crucially important to know how
+AI tools were used.
 
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
+Contributions that used AI to generate significant portions of code,
+comments, or patch descriptions must include an Assisted-by tag in the
+following format::
 
-Where:
+  Assisted-by: LLM # mandatory brief description of usage
 
-* ``AGENT_NAME`` is the name of the AI tool or framework
-* ``MODEL_VERSION`` is the specific model version used
-* ``[TOOL1] [TOOL2]`` are optional specialized analysis tools used
-  (e.g., coccinelle, sparse, smatch, clang-tidy)
+Similarly, usage of other specialized analysis tools should be
+indicated through an Assisted-by tag in the following format::
 
-Basic development tools (git, gcc, make, editors) should not be listed.
+  Assisted-by: [TOOL1] [TOOL2] # optional brief description of usage
 
-Example::
+Examples for ``[TOOL1] [TOOL2]`` are coccinelle, sparse, smatch, and
+clang-tidy. Basic development tools (git, gcc, make, editors) should not
+be listed.
 
-  Assisted-by: Claude:claude-3-opus coccinelle sparse
+Examples::
+
+  Assisted-by: LLM # translate patch description
+  Assisted-by: coccinelle # batch conversion with small manual fixups

davhil01@e142025:~/git/linux$

Signature · 3 lines
Cheers,

David
#51 lore raw
Quoted 51 lines · David Hildenbrand (Arm)
Quoted 39 lines · Laurent Pinchart
Quoted 38 lines · David Hildenbrand (Arm)
Quoted 10 lines · Lorenzo Stoakes

I think the silence suggests nobody :)

What we could do is send a patch to simply add a section on adding a comment explaining _what_ used an LLM and strongly suggesting it would be helpful to do so.

That should hopefully be uncontroversial, then later (or alongside that?) we could submit an RFC for making the language move to 'you MUST do this', as that will probably be the more controversial bit.

Maybe we should just spell out about which scenario we care about where we really expect the disclosure.

Sure, it might be more controversial and we could have that in a separate patch.

I was thinking of the following end result:

Documentation/process/coding-assistants.rst +6 -3
index 899f4459c52d2..4c2ab4dfc6da7 100644
@@ -40,20 +40,37 @@ Attribution
 ===========
 
 When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
+helps track the evolving role of AI in the development process. Further,
+for reviewers and maintainers it is also crucially important to know how
+AI tools were used.
 
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
+Contributions that used AI to generate significant portions of code,
+comments, or patch descriptions must include an Assisted-by tag in the
+following format::
 
Quoted 10 lines

Documentation/process/coding-assistants.rst Attribution · line 55

Quoted 6 lines
Documentation/process/coding-assistants.rst +5 -1
-Where (preferred):
+  Assisted-by: LLM # brief description of usage
+
+Or alternatively::
+
+  Assisted-by: AGENT_NAME:MODEL_VERSION # brief description of usage

I would like to drop this one. As I understand, several people objected to including the product name in the git history as it is just free adverising, and nobody objected to dropping it.

Quoted 3 lines

Yeah. I mean, we could go the following direction (dropping the models and encouraging also a description of usage for other tools), but not sure how much pushback we will get:

I haven't heard anyone involved in this mail thread objecting to the concepts you translated into sentences below, so I think it's worth a try. Of course there may be objections from people who are not aware of this very conversation.

Quoted 43 lines
Documentation/process/coding-assistants.rst +17 -11
index 899f4459c52d2..a085a15d59990 100644
@@ -40,20 +40,26 @@ Attribution
 ===========
 
 When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
+helps track the evolving role of AI in the development process. Further,
+for reviewers and maintainers it is also crucially important to know how
+AI tools were used.
 
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
+Contributions that used AI to generate significant portions of code,
+comments, or patch descriptions must include an Assisted-by tag in the
+following format::
 
-Where:
+  Assisted-by: LLM # mandatory brief description of usage
 
-* ``AGENT_NAME`` is the name of the AI tool or framework
-* ``MODEL_VERSION`` is the specific model version used
-* ``[TOOL1] [TOOL2]`` are optional specialized analysis tools used
-  (e.g., coccinelle, sparse, smatch, clang-tidy)
+Similarly, usage of other specialized analysis tools should be
+indicated through an Assisted-by tag in the following format::
 
-Basic development tools (git, gcc, make, editors) should not be listed.
+  Assisted-by: [TOOL1] [TOOL2] # optional brief description of usage
 
-Example::
+Examples for ``[TOOL1] [TOOL2]`` are coccinelle, sparse, smatch, and
+clang-tidy. Basic development tools (git, gcc, make, editors) should not
+be listed.
 
-  Assisted-by: Claude:claude-3-opus coccinelle sparse
+Examples::
+
+  Assisted-by: LLM # translate patch description
+  Assisted-by: coccinelle # batch conversion with small manual fixups

davhil01@e142025:~/git/linux$

Signature · 3 lines
Regards,

Laurent Pinchart
#52 lore raw
Quoted 4 lines · Laurent Pinchart

I haven't heard anyone involved in this mail thread objecting to the concepts you translated into sentences below, so I think it's worth a try. Of course there may be objections from people who are not aware of this very conversation.

To risk repeating myself - why does it all have to be in one?

Send the non-controversial bit, get that merged so it's there, then follow it up with this. Simple no? :)

Sending this, getting it rejected, then having people's backs up because they'll tie the two things together seems the less politically astute move.

Thanks, Lorenzo

#53 lore raw
Quoted 10 lines · Lorenzo Stoakes
Quoted 4 lines · Laurent Pinchart

I haven't heard anyone involved in this mail thread objecting to the concepts you translated into sentences below, so I think it's worth a try. Of course there may be objections from people who are not aware of this very conversation.

To risk repeating myself - why does it all have to be in one?

Send the non-controversial bit, get that merged so it's there, then follow it up with this. Simple no? :)

Sure, but I don't know what part would be the most controversial :-) I haven't heard any objection in this mail thread.

Quoted 2 lines

Sending this, getting it rejected, then having people's backs up because they'll tie the two things together seems the less politically astute move.

Signature · 3 lines
Regards,

Laurent Pinchart
#54 lore raw
In reply to Laurent Pinchart
Quoted 52 lines · Laurent Pinchart
Quoted 51 lines · David Hildenbrand (Arm)
Quoted 49 lines · Laurent Pinchart
Quoted 34 lines · David Hildenbrand (Arm)
Quoted 10 lines · Lorenzo Stoakes

I think the silence suggests nobody :)

What we could do is send a patch to simply add a section on adding a comment explaining _what_ used an LLM and strongly suggesting it would be helpful to do so.

That should hopefully be uncontroversial, then later (or alongside that?) we could submit an RFC for making the language move to 'you MUST do this', as that will probably be the more controversial bit.

Maybe we should just spell out about which scenario we care about where we really expect the disclosure.

Sure, it might be more controversial and we could have that in a separate patch.

I was thinking of the following end result:

Documentation/process/coding-assistants.rst +3 -3
index 899f4459c52d2..4c2ab4dfc6da7 100644
@@ -40,20 +40,37 @@ Attribution
 ===========
 
 When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
+helps track the evolving role of AI in the development process. Further,
+for reviewers and maintainers it is also crucially important to know how
+AI tools were used.
 
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]

Documentation/process/coding-assistants.rst Attribution · line 55

Quoted 10 lines
Documentation/process/coding-assistants.rst +8 -1
+Contributions that used AI to generate significant portions of code,
+comments, or patch descriptions must include an Assisted-by tag in the
+following format::
 
-Where (preferred):
+  Assisted-by: LLM # brief description of usage
+
+Or alternatively::
+
+  Assisted-by: AGENT_NAME:MODEL_VERSION # brief description of usage

I would like to drop this one. As I understand, several people objected to including the product name in the git history as it is just free adverising, and nobody objected to dropping it.

Quoted 8 lines
Quoted 3 lines

Yeah. I mean, we could go the following direction (dropping the models and encouraging also a description of usage for other tools), but not sure how much pushback we will get:

I haven't heard anyone involved in this mail thread objecting to the concepts you translated into sentences below, so I think it's worth a try. Of course there may be objections from people who are not aware of this very conversation.

Then someone can basically just merge the original patch I sent since everyone seems to agree on that front. And then the extension to include how the LLM was used can be discussed in a separate patch.

#55 lore raw
Quoted 57 lines · Christian Brauner
Quoted 56 lines · Laurent Pinchart
Quoted 54 lines · David Hildenbrand (Arm)
Quoted 39 lines · Laurent Pinchart
Quoted 38 lines · David Hildenbrand (Arm)
Quoted 10 lines · Lorenzo Stoakes

I think the silence suggests nobody :)

What we could do is send a patch to simply add a section on adding a comment explaining _what_ used an LLM and strongly suggesting it would be helpful to do so.

That should hopefully be uncontroversial, then later (or alongside that?) we could submit an RFC for making the language move to 'you MUST do this', as that will probably be the more controversial bit.

Maybe we should just spell out about which scenario we care about where we really expect the disclosure.

Sure, it might be more controversial and we could have that in a separate patch.

I was thinking of the following end result:

Documentation/process/coding-assistants.rst +6 -3
index 899f4459c52d2..4c2ab4dfc6da7 100644
@@ -40,20 +40,37 @@ Attribution
 ===========
 
 When AI tools contribute to kernel development, proper attribution
-helps track the evolving role of AI in the development process.
-Contributions should include an Assisted-by tag in the following format::
+helps track the evolving role of AI in the development process. Further,
+for reviewers and maintainers it is also crucially important to know how
+AI tools were used.
 
-  Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]
+Contributions that used AI to generate significant portions of code,
+comments, or patch descriptions must include an Assisted-by tag in the
+following format::
 
Quoted 10 lines

Documentation/process/coding-assistants.rst Attribution · line 55

Quoted 6 lines
Documentation/process/coding-assistants.rst +5 -1
-Where (preferred):
+  Assisted-by: LLM # brief description of usage
+
+Or alternatively::
+
+  Assisted-by: AGENT_NAME:MODEL_VERSION # brief description of usage

I would like to drop this one. As I understand, several people objected to including the product name in the git history as it is just free adverising, and nobody objected to dropping it.

Yeah. I mean, we could go the following direction (dropping the models and encouraging also a description of usage for other tools), but not sure how much pushback we will get:

Quoted 8 lines
Quoted 4 lines

I haven't heard anyone involved in this mail thread objecting to the concepts you translated into sentences below, so I think it's worth a try. Of course there may be objections from people who are not aware of this very conversation.

Then someone can basically just merge the original patch I sent since everyone seems to agree on that front. And then the extension to include how the LLM was used can be discussed in a separate patch.

https://patch.msgid.link/[email protected]

fwiw

maintainer
#56 lore raw
Quoted 5 lines · Christian Brauner
Quoted 3 lines

Then someone can basically just merge the original patch I sent since everyone seems to agree on that front. And then the extension to include how the LLM was used can be discussed in a separate patch.

https://patch.msgid.link/[email protected]

I still have it around :) I can apply it if that's truly the consensus... I'll confess to having not entirely kept up with the conversation over the holiday weekend, though, and needing to review things.

If you object, do let me know.

Thanks,

jon

#57 lore raw
Quoted 6 lines · Jonathan Corbet
Quoted 4 lines · Christian Brauner
Quoted 3 lines

Then someone can basically just merge the original patch I sent since everyone seems to agree on that front. And then the extension to include how the LLM was used can be discussed in a separate patch.

Quoted 8 lines
Quoted 1 line

https://patch.msgid.link/[email protected]

I still have it around :) I can apply it if that's truly the consensus... I'll confess to having not entirely kept up with the conversation over the holiday weekend, though, and needing to review things.

If you object, do let me know.

I think we need the commit message to be a cleaned up a little :)

Also good to have it out there non-RFC so people know it's actually a contender for merge.

So I think Christian should send a fresh non-RFC version of that with an updated commit msg + with the people here cc'd, and I'm fine with that patch going in as it is (I already gave a tag).

We can always follw up with the comment stuff later (which I consider _really_ important but one thing at a time).

Quoted 4 lines

Thanks,

jon

Cheers, Lorenzo

#58 lore raw
In reply to Jonathan Corbet
Quoted 6 lines · Jonathan Corbet
Quoted 4 lines · Christian Brauner
Quoted 3 lines

Then someone can basically just merge the original patch I sent since everyone seems to agree on that front. And then the extension to include how the LLM was used can be discussed in a separate patch.

Quoted 6 lines
Quoted 1 line

https://patch.msgid.link/[email protected]

I still have it around :) I can apply it if that's truly the consensus... I'll confess to having not entirely kept up with the conversation over the holiday weekend, though, and needing to review things.

I read the article. :) Feel free to massage the commit message to whatever seems right and resubmit.

#59 lore raw
Quoted 9 lines · Christian Brauner
Quoted 8 lines · Jonathan Corbet
Quoted 5 lines · Christian Brauner
Quoted 3 lines

Then someone can basically just merge the original patch I sent since everyone seems to agree on that front. And then the extension to include how the LLM was used can be discussed in a separate patch.

https://patch.msgid.link/[email protected]

Quoted 8 lines
Quoted 4 lines

I still have it around :) I can apply it if that's truly the consensus... I'll confess to having not entirely kept up with the conversation over the holiday weekend, though, and needing to review things.

I read the article. :) Feel free to massage the commit message to whatever seems right and resubmit.

It probably got lost at this point so also please add my R-b to it:

Cheers, Lorenzo

author Reviewed-by
#60 lore raw
Quoted 15 lines · Lorenzo Stoakes (ARM)
Quoted 14 lines · Christian Brauner
Quoted 12 lines · Jonathan Corbet
Quoted 5 lines · Christian Brauner
Quoted 3 lines

Then someone can basically just merge the original patch I sent since everyone seems to agree on that front. And then the extension to include how the LLM was used can be discussed in a separate patch.

https://patch.msgid.link/[email protected]

I still have it around :) I can apply it if that's truly the consensus... I'll confess to having not entirely kept up with the conversation over the holiday weekend, though, and needing to review things.

Quoted 7 lines
Quoted 3 lines

I read the article. :) Feel free to massage the commit message to whatever seems right and resubmit.

It probably got lost at this point so also please add my R-b to it:

Ditto:

Reviewed-by
#61 lore raw
In reply to Lorenzo Stoakes (ARM)
Quoted 23 lines

Op 22-07-2026 14:37 CEST schreef Lorenzo Stoakes (ARM) <[email protected]>:

Quoted 9 lines · Christian Brauner
Quoted 8 lines · Jonathan Corbet
Quoted 5 lines · Christian Brauner
Quoted 3 lines

Then someone can basically just merge the original patch I sent since everyone seems to agree on that front. And then the extension to include how the LLM was used can be discussed in a separate patch.

https://patch.msgid.link/[email protected]

Quoted 8 lines
Quoted 4 lines

I still have it around :) I can apply it if that's truly the consensus... I'll confess to having not entirely kept up with the conversation over the holiday weekend, though, and needing to review things.

I read the article. :) Feel free to massage the commit message to whatever seems right and resubmit.

Quoted 5 lines

It probably got lost at this point so also please add my R-b to it:

Cheers, Lorenzo

I think there is broad support for this, but might it be better to await a more comprehensive change after the maintainer summit?

Anyhow,

#62 lore raw
In reply to Christian Brauner
Quoted 9 lines · Christian Brauner
Quoted 8 lines · Jonathan Corbet
Quoted 5 lines · Christian Brauner
Quoted 3 lines

Then someone can basically just merge the original patch I sent since everyone seems to agree on that front. And then the extension to include how the LLM was used can be discussed in a separate patch.

https://patch.msgid.link/[email protected]

Quoted 7 lines
Quoted 4 lines

I still have it around :) I can apply it if that's truly the consensus... I'll confess to having not entirely kept up with the conversation over the holiday weekend, though, and needing to review things.

I read the article. :) Feel free to massage the commit message to whatever seems right and resubmit.

Oh, one nit with this change...

checkpatch.pl complains about the Assisted-by: format being wrong. Can we also fix that to accept the new format?

Signature · 1 line
Jeff Layton <[email protected]>
#63 lore raw
In reply to Lorenzo Stoakes
Quoted 76 lines

Op 07-07-2026 11:20 CEST schreef Lorenzo Stoakes <[email protected]>:

Quoted 49 lines · David Hildenbrand (Arm)
Quoted 48 lines · Laurent Pinchart
Quoted 27 lines · Lorenzo Stoakes

On Fri, Jul 03, 2026 at 09:05:58AM +0200, David Hildenbrand (Arm) wrote:

I personally think 1-2 are the only relevant cases.

Quoted 10 lines

Yes, I raised something similar as reply to Christian's RFC [1], where I said that for me the information *how* it was used is much more important:

"

Assisted-by LLM # translate commit message
Assisted-by LLM # generate some test cases
Assisted-by LLM # cleanup logic
Assisted-by LLM # everything and I have no clue what any in here does

"

Yup, and we don't need complicated rules for that just 'document what you used it for and give a sense of how much'.

It's fuzzy but useful.

Quoted 2 lines

That tag is it stands is pretty useless, really.

Not to go over it all again but I disagree, even as it stands, it allows us to engage in conversation about the LLM usage if admitted, and to point those who are misbehaving at the rules if not.

Quoted 8 lines

And it is a clear way to get the boolean 'is this person saying they used an LLM'.

But I agree with you it'd be MUCH more useful if we did the above.

I wonder if we could get consensus on adding a section to the doc saying that it'd be _useful_ to add a comment explaining _what_ you did, and explaining the concept with some examples?

I'd support a patch that replaces

Assisted-by Claude:claude-3-opus coccinelle sparse

with

Assisted-by LLM # generate some test cases

and rewrites the Attribution section of Documentation/process/coding-assistants.rst accordingly.

Quoted 8 lines
Quoted 5 lines

I think most people in this mail thread have expressed that how generative AI was used is the most important information, and several people (including myself) have expressed a desire to stop the free advertising. Unless I missed something, I don't think anyone has expressed an interest in keeping the agent name and model.

Yes, that's something I would enjoy seeing. Who reading along has a problem with that? (assuming that the tag must stay)

I think the silence suggests nobody :)

What we could do is send a patch to simply add a section on adding a comment explaining _what_ used an LLM and strongly suggesting it would be helpful to do so.

That should hopefully be uncontroversial, then later (or alongside that?) we could submit an RFC for making the language move to 'you MUST do this', as that will probably be the more controversial bit.

I don't see any reason why we can't just send the softer version right away.

Quoted 1 line
Quoted 6 lines
Quoted 4 lines

--
Cheers,

David

Cheers, Lorenzo

I haven't really seen anyone opposed the following I wrote earlier:

- it's more relevant to know how an LLM was used than that it was used (and the tag is just a quick indication of this)
- advertising for particular corporations does not really serve any purpose to the community, so just say "LLM"
- the tags may be used to prevent arguing
- the requirement to be open about LLM use (whether by tags or whatever) makes it easier for maintainers (if they feel the need) to de-prioritize patches by someone unwilling to abide by these rules whenever the maintainer perceives or suspects prior undeclared LLM use. This prevents an asymmetrical situation of LLM slop being produced much faster that what can be reviewed.

Perhaps the last point is more controversial.

I would really like to adept the systemd policy text.[1] If I have time later today I'll try to write something up... although maybe it is too ambitious :)

[1]: https://lore.kernel.org/linux-fsdevel/20260702-bahnen-ertappen-verspannungen-0eaaf1e3f5af@brauner/

#64 lore raw
Quoted 13 lines · Jori Koolstra

I haven't really seen anyone opposed the following I wrote earlier:

- it's more relevant to know how an LLM was used than that it was used (and the tag is just a quick indication of this)
- advertising for particular corporations does not really serve any purpose to the community, so just say "LLM"
- the tags may be used to prevent arguing
- the requirement to be open about LLM use (whether by tags or whatever) makes it easier for maintainers (if they feel the need) to de-prioritize patches by someone unwilling to abide by these rules whenever the maintainer perceives or suspects prior undeclared LLM use. This prevents an asymmetrical situation of LLM slop being produced much faster that what can be reviewed.

Yup completely agree.

I just want to get something merged rather than just talk :) and so to avoid any _possible_ argument 'keep it like it is but add a comment' is a much, much easier sell to everybody.

Then we can follow up with anything else one we get _something useful merged_ :)

Quoted 2 lines

Perhaps the last point is more controversial.

Not in my opinion. We're only at the beginning of this horror show and I think far stronger measures will eventually be needed (e.g. newcomes will have to build trust first, which sucks, but it may be the only practical solution).

Quoted 5 lines

I would really like to adept the systemd policy text.[1] If I have time later today I'll try to write something up... although maybe it is too ambitious :)

[1]: https://lore.kernel.org/linux-fsdevel/20260702-bahnen-ertappen-verspannungen-0eaaf1e3f5af@brauner/

:)

Cheers, Lorenzo

#65 lore raw
In reply to Lorenzo Stoakes
Quoted 6 lines
Quoted 2 lines

That tag is it stands is pretty useless, really.

Not to go over it all again but I disagree, even as it stands, it allows us to engage in conversation about the LLM usage if admitted, and to point those who are misbehaving at the rules if not.

Just to comment on that

When I don't see the tag I am clueless *if and how* LLMs were used, in particular as some subsystems seem to encourage not using them and we even document it as a "rather nice to have" -- "should".

And if I see the tag, I am clueless *how* LLMs were used.

So I see your point that the tag as it stands helps a little bit, for my day-to-day work, however, not much, as I really care about the "how". :(

Signature · 3 lines
Cheers,

David
#66 lore raw
In reply to Lorenzo Stoakes
Quoted 49 lines · On Thu, 2 Jul 2026 17:50:15 +0100 Lorenzo Stoakes <[email protected]> wrote:
Quoted 48 lines · Jeff Layton
Quoted 30 lines · Greg KH
Quoted 29 lines · Laurent Pinchart
Quoted 27 lines · Jeff Layton
Quoted 12 lines · Greg KH
Quoted 6 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

Then let's move to get people to follow it.

Quoted 1 line

2/ It basically serves as free advertising for proprietary LLM companies.

Quoted 8 lines

Who cares, make up a name, all I want is the "signal" that someone is using a LLM so that I can review it as-such. And if I think someone is not reporting that, I can ask for them to properly attribute it and if they lie, well, that's on them.

Quoted 1 line

3/ It's not clear why we want to collect this info in the first place.

We want to know if a LLM is being used.

But why? What do you intend to do with this information?

Do you mean to use it as an indicator that the patch should receive "extra" review (or maybe that it should be ignored)? Do you mean to use it to generate some sort of statistics at a later time?

Quoted 9 lines
Quoted 3 lines

I use the information to decide how to review the patch, and what level of priority to give it. For that usage I don't need a tag, but I need the information in some human-readable form at patch submission time.

Same here. I don't care about stats, I care about "how do I review this patch" and this gives me that signal that I need if faced with a llm-helped patch.

Do we need a tag for this though?

This seems like the kind of information that we would always require in the cover letter of a series (or the little place in an individual patch for comments that don't get merged). That would also allow you to convey a lot more nuance about how it was used.

Quoted 4 lines
Quoted 4 lines

ISTM asking people to disclose LLM usage in a cover letter would give everyone what they want: Information about whether and possibly how an LLM was used, and it also wouldn't clutter up the changelogs with these tags.

I understand the concern is having the information in the git log. I agree in a level. I wish cloning linux takes not too long time and huge traffic. I wish each commit message be lengthy only as much as needed.

Quoted 4 lines

It's much much clearer and easier to just have a standardised tag for that.

You can see that (and grep for that) immediately, vague paragraphs not so much.

I also agree having a formal tag can be convenient. LWN's Assisted-by: tag statistic was interesting enough to make me supporting this idea.

What about keep requesting this formal signature to the submitters, but also formally allow (or, encourage) subsystem maintainers to remove the tags when pull-request? That's already allowed, to my understanding and some maintainers
are doing [1] today. That can help preventing the git history of a subsystem
bloating or advertising priorietary companies, as much as the maintainer care. Meanwhile, we can still get the information at review time, and also after review from the mail archives. It will still be handy as long as we use tools like 'lei'. As some maintainers are already removing the tags from the git log, this may be anyway a better way. Someone may interested in only their subsystem history, but I find LLM users tend to contribute to multiple subsystems, so this might actually matters.

It might be too clear to formally document this, though.

[1] https://lore.kernel.org/[email protected]

Thanks,
SJ

[...]

#67 lore raw
In reply to Jeff Layton
Quoted 15 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

2/ It basically serves as free advertising for proprietary LLM companies.

3/ It's not clear why we want to collect this info in the first place.

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

Quoted 6 lines

---
Christian had proposed watering down the LLM attribution, but I think it's not productive to try and track this until we have a clearer sense of what we want to do with this information and how to make it more reliable.

I agree that the current tagging system is flawed and almost useless for real analysis -- self-reporting and a nebulous definition of what "LLM was used" actually means are red flags for any data scientist.

I don't have a stand on whether the tagging should be removed or fixed. But today I ignore it (my 2-cents US) for these reasons.

Signature · 1 line
Chuck Lever
#68 lore raw
Quoted 25 lines · Chuck Lever
Quoted 15 lines · Jeff Layton

We've had this requirement in place in the Documentation for several months, but it's becoming clear that the signal to noise ratio from this is quite low.

1/ It's not universally followed. While many people do try to attribute the LLMs in good faith, not everyone does for various reasons.

2/ It basically serves as free advertising for proprietary LLM companies.

3/ It's not clear why we want to collect this info in the first place.

Given that the data this provides is flawed at best and is being collected for a purpose that isn't clear, let's just kill the requirement for these tags from the kernel at large.

Quoted 6 lines

---
Christian had proposed watering down the LLM attribution, but I think it's not productive to try and track this until we have a clearer sense of what we want to do with this information and how to make it more reliable.

Quoted 6 lines

I agree that the current tagging system is flawed and almost useless for real analysis -- self-reporting and a nebulous definition of what "LLM was used" actually means are red flags for any data scientist.

I don't have a stand on whether the tagging should be removed or fixed. But today I ignore it (my 2-cents US) for these reasons.

Oh no doubt they have highly dubious value as _data_.

But they are useful for practical purposes :) i.e. 'ok I can talk about the LLM-ness here without it being quite so aggressive to do so because it's admitted'.

And also political value in that 'here is how you are supposed to do it' vs. <how you did it>.

I think there's definite practical human stuff that they provide sufficient for there to be value here, and I've found that in practice also :)

(Though of course it's far far from resolving how we deal with the unattributed stuff, which remains an ongoing issue!)

Quoted 3 lines

--
Chuck Lever

Cheers, Lorenzo