Pasting a huge AI generated explanation to a problem in an issue or pull-request is nothing but RUDE. Don't do it. You look stupid and the receivers of that feel insulted.
We are humans. We communicate like humans. Fine, use the tools you like, but don't insult us.
@safigo @mk I'll answer literally since your "why?" seems well-intentioned, given your other comments.
Painting a choice between "AI slop description" and "no description at all" is a false dicothomy, which is a rhetorical fallacy ill-intentioned people try to pull: portraying a bad-thing-X as "X or even-worse-Y". It's a fallacy because those are not the only options. The issue/PR author always has the 3rd option of writing a normal human message. That's why you got the "false dilemma" reply.
I did this Summary A signed integer overflow in tftp_set_timeouts caused by Curl_timeleft_ms() returning TIMEDIFF_T_MAX (9223372036854775807), which when added to 500 at tftp.c:171 overflows the ti...
@bagder now I see. You were too polite in your description. I thought it would be a PR with a description generated by LLM.
Your cases are just copy paste from a LLM after asking it to find a security issue, without understanding or even not reading an answer…
It’s sad…
@bagder @safigo Raymond Chen calls these "being on the other side of an airtight hatchway", which I really like.
Sorry you are dealing with this especially when so overworked from recent security reports. 
https://devblogs.microsoft.com/oldnewthing/20221004-00/?p=107246
@kumarvibe I guess I would consider "contains irrelevant minutiae" as "inaccurate"
I don't think "devoid of conversational inflection" is a problem
@kumarvibe Sorry for repeated posts, I'm having difficulty putting what I'm trying to get at into words. Maybe something like:
> The point is, don't make the maintainer wade through output to get to the point. If you didn't want to read it, the maintainer probably doesn't want to either.
In vcpkg we get lots of submissions that "look" like they're AI that are fine but we also get slop. I would not want to encourage changing the submissions that are good into having the submitter restate what the LLM said but worse, or omit bits that are important justification. But I would want them to prune the AI output of irrelevant slop garbage.
@malwareminigun @bagder I think I would say that even then, if an LLM response is unsolicited it’s unwelcome.
If I ask someone a question and they respond by pasting an LLM response, what contribution have they made? If I wanted to ask an LLM I could just do it myself; if I ask a human it’s because I want to know their answer. (Feels similar to obnoxious ‘let me google that for you’ answers, in a way.)
I suppose what I’m wondering is, what are the situations where responding with LLM output is *not* crappy conduit to chatbot behaviour as you say?
@benjamineskola That's the difficulty; it's one of those "I know it when I see it" situations.
I've seen clear LLM responses describing, for example, why a build fails or does something incorrectly included in PR descriptions, and those have been reasonable. Often the tools are better at citing locations for information than the average contributor which can make reviewing those contributions easier.
It's about the contributor understanding what they're doing, acting with intent, expecting accountability for the results even if an LLM they're using produced them, and having respect for maintainers' time.
Just pasting what a maintainer asks into the LLM and then uncritically pasting what the LLM says back is unacceptable.
Making what could be said in a 1 line PR message into paragraphs describing every individual semicolon in the change is unacceptable.
The contributor not having read their own submission is unacceptable.
Not disclosing which parts of a contribution are produced by an LLM is unacceptable.
But I don't think that makes all LLM generated content unacceptable. "I'm making this change because GPT 5.4 said <blah>" can be fine as long as <blah> is correct and not full of irrelevant garbage.
i love how llms are good at finding bugs, but not actually providing a fix for them.
the one thing an automated llm would be useful for it doesn't actually do lmao
(god... imagine not having to spend hours fixing bugs and security vulns in the code)
@bagder "Not worth writing, not worth reading" has a delightfully similar cadence to "not my circus, not my monkeys", and in the last year or two it's given me about the same peace of mind.
Of course, for vuln reports, you have to be less picky about who's reporting and what tools they use, as long as the bug is real. A bug's a bug. But in a context where hundreds of people are throwing big token budgets at a single project and largely reporting duplicates, I'd guess the odds of missing something important from a reporter who didn't even bother to write the ticket (let alone do their own dupe-checking diligence) are pretty sparse.
@bboreham.bsky.social The problem is you don't know if it is slop copy-paste or a genuine human response because — and it might be a shock to a lot of people who love TL:DR's and hate long one-piece messages instead of 50 one-sentence messages — people actually are capable of writing more than 10 words when they want to be understood correctly or to show their logic behind their answer. Also things are often complex and cannot be simplified down to a catchphrase.
Redis or Memcashed (example from the link)? I don't know the subject, I don't even know if it is entirely fictional for the purpose of example but I certainly know that the answer never will be just Redis or just Memcashed. Because it depends. Because people can have opinions. With a logic behind those opinions, how they arrived to stand with Redis/Memcashed. And logic — this is the bit that is important and cannot be simplified.
Also, AI style isn't the unique AI style. It is result of training. On human text.
I, myself, am AI sceptic. I don't use it, the more I hear about it the more reasons I have to hate it. This "no slop grenade" shit — is one of them. We can't even trust if a person wrote their genuine thoughts to you or if they just ctrl+c ctrl+v AI slop to you.
Oh, it is long? It has structure? It is generally grammatically correct? Welp, that's certainly AI because, quote: "Nobody writes essays in Slack. It's only possible because of AI copy-paste."
I call bullshit. I write essays in chat. All the time. Been doing it my whole life. And now, instead of "that was great explanation" or "thank you for clarity" I will get "here is no slop grenade, begone"? No in hell, I won't!
@bboreham.bsky.social and honestly, I don't know what to do with it. If we will punish those, who clearly state "I asked AI here is a copy paste" (and that kind of people thankfully still exist) — they simply will stop... leading with "I asked AI here is a copy paste" and just will copy paste without any indication whatsoever. And then you are stuck with questioning "is it AI or not?" each time you see a post longer than 20 words.
And noslopgrenade.com won't save you. People copy pasting AI responses will not understand your demand "not AI please". Especially in such a passive-aggressive manner as "here is a link with term 'slop' in it to a whole site dedicated to roasting what you just did with your AI thing because I can't be bothered to even do it myself". They will not understand. They wanted to help. They got from conversation, talked with the thing, sent result to you. They think they did work.
And this 3-posts long rumbling is a proof not only AI can write long.
@Lemonid there is skill involved in knowing whether to use it, that doesn't inherently make the tool useless or ill-conceived. "Batteries not included", bring your own senses for that.
The problem with the provided example is actually outlined on the page quite well: response from an LLM was generic and did not account for the implicit context of their jobs and what they're working on. Such a response implies, as written, that the sender cannot be bothered to do general research before asking another person for specifics in an act of disrespect for their time.
It's not the style. It's the substance. Or lack thereof, more specifically.
I pity the company culture of whoever gets to use this with enough of a frequency to justify building such a reusable response, but having worked for just under a decade in corporate before the rise of LLMs I am certain that some people will provide good cause for a response like this. It saddens me, but does not surprise at all.
@bagder At work, I had a report with data I pulled every week for the past year. Beautifully streamlined and perfectly easy to understand. An idiot manager asked for a report on exactly what I had. I sent it to him. He fed it to an AI which hallucinated most of the data incorrectly and turned my charts into a jittery slop mess. He said, thanks, I'll use the report I got from Claude.
You mean the report I spent a year meticulously updating before you fed it into a woodchipper?
Yes the llm written ones gcc is getting is annoying. The descriptions are just plain wrong. Lucky the testcases are real ones. And some are annotated just badly. Lucky we can remove the comments and get to the real issue and ignore the llm written parts.
Fair. That link you posted was insane.
I have work requirements regarding spicy autocorrect. So I'll write the code, then have AI review the changes for an initial bullet point list of PR changes for the commit. It's not bad, but sometimes thinks that 'for' loop was really important.
daniel:// stenberg://
Anton Ohorodnyk
Michael Knudsen
Hisham
Billy O'Neal
Sharkfie 
Sibshops
F4GRX Sébastien
Ondrej Moravek
Piko Starsider 
NeonkAaa
Lincoln Stoll
Kumar McMillan
ben
numanumayey
Josh Law
MaddieM4
Sam Stephens
FynnND
Justin Fitzsimmons
Bryan Boreham
Cytrynid
maxotaf
D:\side\>
Sean O'Brien
Kaito
luca0N!
MintyFries
L
Stephanie
Yannis G
@iveyline
Frederic Thevenet
m_eiman
happyborg
Nuclear Oatmeal 
Mage of Voids
Don Armstrong
Gérard Fadrelle