Sigh, the AI Slop has now come for the PHP project.

[derickr@phpc.social]

Sigh, the AI Slop has now come for the PHP project.

https://github.com/php/php-src/pull/21317

#AiSlop #php

[afilina@phpc.social]

@derickr A 10,000-line PR by a bot, for me, is 100% unmergeable. It
s basically a black box. No way someone can diligently review this and be sure that it covers everything, while considering all the impacts.

[jrf_nl@phpc.social]

@derickr I'm surprised and disappointed that the PR is still open and people are seemingly engaging with it.

The copyright and license question (what code was this AI trained on and what was the license of that code) alone should mean the PR should be closed and locked.

[girgias@phpc.social]

@jrf_nl @derickr as with always with PHP no one has the "authority" to do anything. And doing anything that is "drastic" or "radical" (like banning a user that routinely provides bad advices, refuses to follow procedures, etc.) is met with being told that you are being the issue rather than them.

[dbu@phpc.social]

@Girgias @jrf_nl @derickr what surprises me is that it is it not ignored until a proper RFC is voted on. Why discuss implementation details (of an llm) instead of discussing the RFC first?

[maccath@phpc.social]

@dbu @Girgias @jrf_nl @derickr This. I know it's a draft PR, but it didn't need to be submitted at all in its current state. They can iterate in their own fork. If I were moderating, I'd personally close with 'needs RFC' to avoid wasting people's time.

[jrf_nl@phpc.social]

@maccath @dbu @Girgias @derickr That's a completely different argument though - you are accepting LLM use, but not the noise of a PR without an RFC.

I'm suggesting not accepting LLM use as it is all based on theft of copyrighted works and does not respect the license of the code it was trained on.

[dolanor@hachyderm.io]

@dbu @Girgias @jrf_nl @derickr and also the identity of the account. The person wants to keep its identity private.

Like WTF?
You're doing a drive by PR anonymously and you want people to merge your AI generated shit?
What about the IP?

Yeah, it's crazy the people from PHP leave that opened.

[timwolla@phpc.social]

@dbu @Girgias @jrf_nl @derickr An RFC without a reasonable (PoC) implementation is worthless, especially so for generics, which are well understood to be “desired”.

I suspect you would have no issues getting a “generics” RFC accepted, the hard part has always been the implementation.

[timwolla@phpc.social]

@dbu @Girgias @jrf_nl @derickr When taking a look at the implementation to determine if it's any good, then adding a handful of comments with nits (as I did) is not hard and will make sure that any future changes and fixes will take the nits into account, ultimately saving everyone a little bit of effort.

[timwolla@phpc.social]

@maccath @dbu @Girgias @jrf_nl @derickr It has explicitly been stated in the PR that an RFC is require for it to go anywhere (including labeling it as such).

I expect anyone that is (still) engaging with the PR to do it in full knowledge of the procedures. The handful of remarks made there would equally (need to) be made as part of the RFC discussion to determine the desired semantics.

[loptimist@piaille.fr]

@derickr Even more concerning, it seems that members of the PHP org are actively supporting that initiative. Seems very suspicious.

[luciledt@phpc.social]

@derickr I think the worst is just how the person maintaining the PR is just feeding each remark to their LLM and throwing at reviewers the output. Some comments are quite useful and try to make the author think about their work but it really feels like they can't do it themselves at all and need their bot to do all the work...

[dbu@phpc.social]

@timwolla fair enough. And of course actual generics with a good standard and clean implementation would be fantastic to have.
Which leaves the fundamental question about llm code and copyright - personally I fear that ship has sailed a while ago.

[jrf_nl@phpc.social]

@timwolla Reviewing a PR which should never have been created, let alone considered, is the same as "feeding the trolls".

It also leaves the impression that:
a) the PHP project is seriously considering the code for inclusion (otherwise, why review?)
b) creating a PoC for a feature using an LLM is acceptable in the PHP project (which it shouldn't be IMO - PoCs too often end of up in a project instead of being thrown away and it legitimizes theft)

/cc @maccath @dbu @Girgias @derickr

[derickr@phpc.social]

@timwolla @maccath @dbu @Girgias @jrf_nl I find the feeding of PHP dev's remarks directly into their chat bot an extremely rude thing to do. Each interaction also sends everybody involved a notification, and having to sift through AI slop on a Sunday morning/afternoon is not something any sane person should be exposed to.

[timwolla@phpc.social]

@derickr @maccath @dbu @Girgias @jrf_nl I agree on using AI for the discussion, if I wanted to talk to a computer I'd do it myself.

With regard to notifications it's unfortunate that the target branch change requested reviews from every code owner (but that is a common mistake for contributors to make).

Folks that are not or no longer interested in following the conversation further can just unsubscribe from the PR.

[phoopee3@phpc.social]

@luciledt @derickr I hate the the person(?) eventually is like “this is just a proof of concept” but like, you’re wasting a ton of people’s real time on your poc and they didn’t ask for that.

[ondrej@mastodon.rfc1925.org]

@derickr I will not disclose my name yet. What the actual fuck?

[ramsey@phpc.social]

@Girgias @jrf_nl @derickr I know someone (not me) who pointed a different agent at it and asked it to find security problems, and that agent suggested that some of the code looked like it might be attempting to poison the opcache.

Of course, I have no idea how to determine whether that’s true because the other agent could be hallucinating, but a massive PR from an unknown entity does concern me that there’s something potentially nefarious buried within.