Conversation
Krzysztof Kozlowski
krzk
Edited yesterday
AI slop in Open Source is not only about receiving poor quality LLM generated junk patches. It is also about receiving replies generated by LLM to reviewer's comments. Such replies are overly long, overly polite, unnecessarily "You are right, <here goes very long explanation>".
Reading an LLM generated answer to my review comment on the mailing list is a waste of my time.
If you are a contributor to an open source project, understand that maintainer has absolutely ZERO interest in talking to your LLM through you.
Reading an LLM generated answer to my review comment on the mailing list is a waste of my time.
If you are a contributor to an open source project, understand that maintainer has absolutely ZERO interest in talking to your LLM through you.
1
11
19
Geert Uytterhoeven
geert@society.oftrolls.com@krzk So it wasn't you who enabled Sashiko for the devicetree mailing list?
1
0
0
@geert For example (but that is not the only one) look at this reply:
https://lore.kernel.org/all/56f5e73b-5f40-4bfb-9796-dadfcb4f9085@oss.qualcomm.com/
which is too long LLM junk "you are right".
And also consider this:
https://lore.kernel.org/all/01578e6a-d10a-46df-bb32-fd45ecb365d7@oss.qualcomm.com/
which is not even touching the subject and not answering my actual comments. It's a perfect LLM answer which is not solving anything, just wasting my time and ticking a box "I need to answer whatever to the reviewer".
https://lore.kernel.org/all/56f5e73b-5f40-4bfb-9796-dadfcb4f9085@oss.qualcomm.com/
which is too long LLM junk "you are right".
And also consider this:
https://lore.kernel.org/all/01578e6a-d10a-46df-bb32-fd45ecb365d7@oss.qualcomm.com/
which is not even touching the subject and not answering my actual comments. It's a perfect LLM answer which is not solving anything, just wasting my time and ticking a box "I need to answer whatever to the reviewer".
1
0
1
Geert Uytterhoeven
geert@society.oftrolls.com@krzk I am not sure these are written by an LLM.
However, both are about the SCMI train wreck ...
1
0
0
Krzysztof Kozlowski
krzk
@geert It is, looks exactly like that. Also AI agrees with that, quoting one LLM analysis:
"Several characteristics are consistent with LLM-generated technical email:
- Repetitive structure and over-explanation:...
- Highly polished corporate tone:...
- Excessive explicit enumeration:...
- Defensive balancing language:
- Commit message rewrite is unusually verbose:..."
Native speakers don't write like that. Non-native even less.
But to be fair, LLM judged that it looks still like technically valid answer, not AI slop:
"That combination often means one of:
- Written by a knowledgeable engineer using AI for drafting/polishing.
- AI-generated from detailed technical notes.
- Human-written but heavily edited with AI assistance."
"Several characteristics are consistent with LLM-generated technical email:
- Repetitive structure and over-explanation:...
- Highly polished corporate tone:...
- Excessive explicit enumeration:...
- Defensive balancing language:
- Commit message rewrite is unusually verbose:..."
Native speakers don't write like that. Non-native even less.
But to be fair, LLM judged that it looks still like technically valid answer, not AI slop:
"That combination often means one of:
- Written by a knowledgeable engineer using AI for drafting/polishing.
- AI-generated from detailed technical notes.
- Human-written but heavily edited with AI assistance."
0
0
0