I think this thread is about whether to restrict LLM-assisted coding in the core tools such as the compiler. @Zoggy proposed, above, adding information to opam packages as to whether they were AI-generated, which had previously been proposed in a different thread. That would just let people who don’t want to incorporate LLM-output into their projects via dependencies know which packages not to use. That could be because of concerns about copyright for legal or ethical reasons or it could because of other ethical concerns, not necessarily because of concerns about slop. It wouldn’t prevent others from using those packages.
I would be another person who wouldn’t want to use such packages because of ethical concerns. I would be unhappy about possibly incorporating LLM outputs into my software via the standard library if it contained them, and I wouldn’t want to endorse or encourage the use of LLMs by my coding choices, such as, e.g., contributing to a project that accepts LLM-assisted contributions, unless it only involved models for which all the ethical problems and environmental sustainability issues were resolved/not present (which might not exist at this time).
5 Likes
Thank you, that was very clarifying. I hadn’t considered the ethical perspective particularly in the opam packages context, and I agree it’s a valid concern. My interpretation was that a policy of that kind (labeling AI-generated packages) would be more restrictive than inclusive in practice, but maybe it wouldn’t be. I understand better the value of being able to distinguish packages authored by humans from the AI-generated ones now.
2 Likes
I’m not going to comment on the advisability of using AI/LLMs. Or on slop. But I -do- feel it’s really, really important that a user of a package be able to know whether the code was of the package was authored by AI/LLM, and perhaps even going further, whether the tests were. Heck, probably the documentation, though I guess I don’t really care much about that. I’m a “use the source, Luke” guy. Also a “sometimes the tests are the best form of documentation” (which is why it’s important to know whether the tests were written using AI/LLM).
Each of us has different criteria, different feelings about AI/LLM, different reasons to accept or reject AI/LLM in various contexts. I don’t want to describe my own, b/c that’s really not relevant in this conversation.
7 Likes
First time hearing about reative, well done. I would def try it out.
is there any place were I can give unorganised feedback, etc?
I had a quick look at reativa and prescriptive. They seem very interesting, and I’d encourage you to announce them in the forum.
The ocaml community is small but diverse, not everyone is rejecting AI categorically. There is an AI channel in the discord for example.
The only instance I can remember of strong AI rejection, was a few months ago when someone was trying to push their unsolicited slop PRs at an already stretched compiler team.