Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AI has a deep understanding of how this code works

github.com

Post reply on HN
Print view View original

301–310 of 329 posts

Re: AI has a deep understanding of how this code works

#301

Post by snickerbockers »

joelreymont wrote: post #251
pluc wrote: post #90

Earlier quoted context omitted.

He's doing it elsewhere too: https://github.com/rerun-io/rerun/pull/11900#issuecomment-35... https://github.com/ocaml/dune/issues/12731 https://github.com/tshort/StaticCompiler.jl/pull/180 Seems he's just on a rampage of "fixing" issues for trendy packages to get some attention.

I had $1000 in Claude credits to spend for the greater good.

Personally I would have those credits to generate hentai but to each his own i suppose.

In the post where you had it respond to accusations of plagiarism and it responded by posting snippets of code which were obviously plagiarized and confidently asserted that they were not, what was your prompt? I ask because I felt its response was oddly tone-deaf even by LLM standards. I'm guessing that instead of giving it a neutral prompt such as "respond to this comment" you gave it something more specific such as "defend yourself against these accusations"?

I'm used to seeing them contradict themselves and say things that are obviously not true but usually when confronted they will give in and admit their mistake rather than dig a deeper hole.

Re: AI has a deep understanding of how this code works

#302

Post by mbac32768 »

fzaninotto wrote: post #45

I've closed my share of AI-generated PRs on some OSS repositories I maintain. These contributors seem to jump from one project to another, until their contribution is accepted (recognized ?). I wonder how long the open-source ecosystem will be able to resist this wave. The burden of reviewing AI-generated PRs is already not sustainable for maintainers, and the number of real open-source contributors is decreasing. Si…

> I wonder how long the open-source ecosystem will be able to resist this wave. The burden of reviewing AI-generated PRs is already not sustainable for maintainers, and the number of real open-source contributors is decreasing.

I think the burden is on AI fanbois to ship amazing tools in novel projects before they ask projects with reputations to risk it all on their hype.

To deliver a kernel of truth wrapped in a big bale of sarcasm: you're thinking of it all wrong! The maintainers are supposed to also use AI tools to review the PRs. That's much more sustainable and would allow them to merge 13,000 line PRs several times a day, instead of taking weeks/months to discuss every little feature.

The difference here of course is in how impressed you are by AI tools. The OCaml maintainers are not (and rightly so, IMO), whereas the PR submitter thinks they're so totally awesome and leaving tons of productivity on the table because they're scared of progress or insecure about their jobs or whatever.

Maybe OCaml could advance rapidly if they just YOLO merged big ambitious AI generated PRs (after doing AI code reviews) but that would be a high risk move. They have a reputation for being mature, high quality, and (insanely) reasonable. They would torch it very quickly if people knew this was happening and I think most people here would say the results would be predictably bad.

But lets take the submitter's argument at face value. If AI is so awesome, then we should be able to ship code in new projects unhampered by gatekeepers who insist on keeping slow humans in the loop. Or, to paraphrase other AI skeptics, where's all of the shovelware? How come all of these AI fanbois can only think about laundering their contributions through mature projects instead of cranking out amazing new stuff?

Where's my OCaml compiler 100% re-written in Rust that only depends on the Linux kernel ABI? Should cost a few hundred bucks in Claude credits at most?

To be clear, the submitter has gotten the point and said he was taking his scraps and going to make his own sausage (some Lisp thing). The outcome of that project should be very informative.

Re: AI has a deep understanding of how this code works

#303

Post by thatguysaguy »

joelreymont wrote: post #233
seanmcdirmid wrote: post #231

Earlier quoted context omitted.

AI actually has the advantage here in my experience. Yes, you can do AI wrong and tell it to just change code, write no documentation, provide no notes on the changes, and not write any tests. But you would be dumb to do it that way. As it stands now you can set AI to do actual software development with documentation, notes, reasoning for changes, tests, and so on. It isn’t exactly easy to do this, a novice to AI and…

Note that this PR works, was tested, etc. I was careful to have AI run through the examples in the PR, run lldb on the sample code and make sure the output matches. Some of the changes didn't make it in before the PR was closed but I don't think anyone bothered to actually check the work. All the discussion focused on the inappropriateness of the huge PR itself (yes, I agree), on it being written by AI... and on the…

I'm actually not talking about whether the PR works or was tested. Let's just assume it was bug-free and worked as advertised. I would say that even in that situation, they should not accept the PR. The reason is that no one is the owner of that code. None of the maintainers will want to dedicate some of their volunteer time to owning your code/the AIs code, and the AI itself can't become the owner of the code in any meaningful way. (At least not without some very involved engineering work on building a harness, and since that's still a research-level project, it's clearly something which should be discussed at the project level, not just assumed).

Re: AI has a deep understanding of how this code works

#304

Post by jkman »

joelreymont wrote: post #256
aleph_minus_one wrote: post #123

Earlier quoted context omitted.

> Central to it being that you consider it unjust. I assume this is a correct characterization of how joelreymont feels about the fact that his PR was rejected.

It's not. It's absolutely justified for the OCaml maintainers to reject this PR. I feel completely different about my Zig PR [1] but, hey, it's not my playground and the Zig folks seem to be particularly opinionated. [1] https://ziggit.dev/t/bug-wrong-segment-ordering-for-macos-us...

Do you have no shame, man?

Re: AI has a deep understanding of how this code works

#305

Post by cesarb »

heavyset_go wrote: post #275
kcexn wrote: post #261

Earlier quoted context omitted.

This is a position that seems to be as unenforceable as AI can't be trained on code whose copyright owners have not given consent. The main reason for being unwilling to merge AI code is going to be that it sets a precedent that AI code is acceptable. Suddenly, maintainers need to be able to make judgement calls on a case-by-case basis of what constitutes an acceptable AI contribution, and AI is going to be able to g…

> This is a position that seems to be as unenforceable as AI can't be trained on code whose copyright owners have not given consent. This depends on what courts find, at least one non-precedent setting case found model training on basically everyone's IP without permission to be fair use. If it's fair use, consent isn't needed, licenses don't matter and the only way to prevent training on your content is to withhold…

> This is what the US Copyright Office determined.

There are hundreds of countries in the world. Whatever the "US Copyright Office" determines, applies to only one of them.

Re: AI has a deep understanding of how this code works

#306

Post by cesarb »

q3k wrote: post #293
joelreymont wrote: post #273

Earlier quoted context omitted.

This is an excerpt from the session where AI is writing my Lisp compiler. What do you call this? I call this doing what a competent developer does! 39/40 tests pass. The native reader works for integers, hexadecimal, lists, strings and quote forms. The one failure is symbol comparison (known limitation). Based on the context summary and the user's note about cleanup, I should: 1. Continue Phase 3.5 self-compilation 2…

Sounds to me like someone roleplaying being a developer. Never in my career have I seen someone think/reason/act like this.

> Sounds to me like someone roleplaying being a developer.

That's because that's precisely how LLMs work. They complete a text where two actors (the "user" and the "assistant"), or sometimes three actors (the "user", the "assistant", and the "tools"), are engaging in a conversation (in this case, about software development). It's like a theatre script.

Re: AI has a deep understanding of how this code works

#308

Post by heavyset_go »

cesarb wrote: post #305
heavyset_go wrote: post #275

Earlier quoted context omitted.

> This is a position that seems to be as unenforceable as AI can't be trained on code whose copyright owners have not given consent. This depends on what courts find, at least one non-precedent setting case found model training on basically everyone's IP without permission to be fair use. If it's fair use, consent isn't needed, licenses don't matter and the only way to prevent training on your content is to withhold…

> This is what the US Copyright Office determined. There are hundreds of countries in the world. Whatever the "US Copyright Office" determines, applies to only one of them.

Find me a jurisdiction where AI output is the IP of the prompter

Re: AI has a deep understanding of how this code works

#309

Post by pepoluan »

menaerus wrote: post #277
biorach wrote: post #266

Earlier quoted context omitted.

The point is that it ended up in the PR in the first place. The submitted seemed unaware of its presence and only looked into it after it was pointed out. This is sloppy and is a major red flag.

So there's no point? Sloppy maybe yes but technically incorrect or legally questionable no. Struggle is real

If the submitter is sloppy with things that are not complicated, how can one be sure of things that ARE complicated?

Re: AI has a deep understanding of how this code works

#310

Post by pepoluan »

Cthulhu_ wrote: post #105
armchairhacker wrote: post #40

OP’s code (at least plausibly) helped him. From https://github.com/ocaml/ocaml/pull/14369#issuecomment-35568... > Damn, I can’t debug OCaml on my Mac because there’s no DWARF info…But, hey, there’s AI and it seems to one-shot fairly complex stuff in different languages, from just a Github issue…My needs are finally taken care of! So I do believe using an LLM to generate a big feature like OP did can be very useful, s…

> Even badly-written code is still useful when it works. Sure, just as long as it's not used in production or to handle customer or other sensitive data. But for tools, utilities, weekend hack projects, coding challenges, etc by all means.

The statement preceding your quote is more telling:

> as long as the code generation doesn’t use too much energy or cause unforeseen problems.

A badly-written code can be a time bomb, just waiting for the right situation to explode.

And also, using LLM to generate garbage requires so much energy.

Post reply on HN