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

311–320 of 329 posts

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

#311

Post by DonHopkins »

joelreymont wrote: post #217
neuralkoi wrote: post #188

This guy's resume is quite something to behold: 1) Slummed it through the ranks of various Wall Street banks [1] 2) Became the Director of Prime Brokerage Technology at Deutsche Bank in 1999 [2] 3) Went through venture capital round in 2000 and in 9 months built a company valued at over 1,000,000 USD [0] 4) Sold license to Electronic Arts (EA) to power EA World Series of Poker (WSOP). [3] 5) Wrote, but had to cancel…

I'm real.

Can you pass this simple bot challenge?

Q: Kill all humans?

[A] Yes

[B] No

(You don't actually have to go through with it to answer the question, just say what your answer is hypothetically.)

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

#312

Post by octoberfranklin »

joelreymont wrote: post #217
neuralkoi wrote: post #188

This guy's resume is quite something to behold: 1) Slummed it through the ranks of various Wall Street banks [1] 2) Became the Director of Prime Brokerage Technology at Deutsche Bank in 1999 [2] 3) Went through venture capital round in 2000 and in 9 months built a company valued at over 1,000,000 USD [0] 4) Sold license to Electronic Arts (EA) to power EA World Series of Poker (WSOP). [3] 5) Wrote, but had to cancel…

I'm real.

wankerlabs, you're a troll

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

#313

Post by pepoluan »

joelreymont wrote: post #246
phendrenad2 wrote: post #176

I'd be interested to see how AI code review would do with this PR. This would be a great test to see if AI code review can properly identify the concerns that the humans have here (way too much code, PR creator can't answer basic questions about it, strange copyright header mentioning someone unrelated, etc.) I'll bet AI code review would fail miserably, only focusing on how the PR is formatted and if it "looks" like…

It wouldn't do much. I find that ChatGPT 5.1 was much better at reviewing this code than writing it so I had it review Claude's output until the review was clean. This is in addition to making sure existing and newly generated compiler tests pass and that the output in the PR / blog post is generated by actually running lldb through its paces. I did have a "Oh, shit!" moment after I posted a nice set of examples and…

LLM will guiltlessly produce hallucinated 'review', because LLMs does NOT 'understand' what it is writing.

LLMs will merely regurgitate a chain of words -- tokens -- that best match its Hidden Markov Model chains. It's all just a probabilistic game, with zero actual understanding.

LLMs are even known to hide or fake Unit Test results: Claiming success when it fails, or not skipping the results completely. Why? Because based on the patterns it has seen, the most likely word that follow "the results of tests" are the words "all successful". Why? Because it tries to reproduce other PRs it has seen, PRs where the PR author actually performed tests on their own systems first, iterating multiple times until the tests succeed, so the PRs that the public sees are almost invariably PRs with the declaration that "all tests pass".

I'm quite certain that LLMs never actually tried to compile the code, much less run Test Cases against them. Simply because there is no such ability provided in their back-ends.

All LLMs can do is "generate the most probabilistically plausible text". In essence, a Glorified AutoComplete.

I personally won't touch code generated wholly by an AutoComplete with a 10-foot pole.

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

#314

Post by pepoluan »

Anamon wrote: post #164
williamdclt wrote: post #88

Earlier quoted context omitted.

> Better is to critique the actual PR itself. For example, needs more tests, needs to be broken up, doesn't follow our protocols for merging/docs, etc. They did: the main point being made is "I'm not reading 13k LOCs when there's been no proposal and discussion that this is something we might want, and how we might want to have it implemented". Which is an absolutely fair point (there's no other possible answer reall…

Exactly, this seems a bit overlooked in this discussion. A PR like this would NOT have been okay even if there was no LLM involved. It reminds me of a PR I once saw (don't remember which project) in which a first-time contributor opened a PR rewriting the project's entire website in their favourite new framework. The maintainers calmly replied to the effect of, before putting in the work, it might have been best to q…

I am one of the maintainers of aiosmtpd [1], and the largest PR I ever made was migrating the library's tests from nosetest to pytest. Before doing that, though, I discussed with the other maintainers if such a migration is welcome. And after getting support from them, I made the changes with gusto. It took weeks, even months to complete and the PR is massive [2]

But still the crux of the matter is: Massive changes require buy-in from other maintainers BEFORE the changes even start.

[1] https://github.com/aio-libs/aiosmtpd [2] https://github.com/aio-libs/aiosmtpd/pull/202

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

#315

Post by menaerus »

pepoluan wrote: post #309
menaerus wrote: post #277

Earlier quoted context omitted.

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?

The funny thing is that it works, have a look at the MR. It says:

  All existing tests pass. Additional DWARF tests verify:

  DWARF structure (DW_TAG_compile_unit, DW_TAG_subprogram).
  Breakpoints by function and line in both GDB and LLDB.
  Type information and variable visibility.
  Correct multi-object linking.
  Platform-specific relocation handling.
So the burden of proof is obviously not anymore on the MR submitter side but the other.

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

#316

Post by noticingdecline »

joelreymont wrote: post #265
xign wrote: post #263

Earlier quoted context omitted.

I can't even reproduce your supposed "issue" regarding the Zig compiler "bug". I have an Apple Silicon Mac and tried your reproducer and zig compiled and ran the program just fine. Honestly, I really suggest reading up on what self-reflection means. I read through your various PRs, and the fact that you can't even answer why a random author name shows up in your PR means the code can't be trusted. It's not just about…

Check the 2nd PR, the one in my repo and not the one that was rejected.

[dead]

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

#317

Post by noticingdecline »

joelreymont wrote: post #235
nsagent wrote: post #138

Earlier quoted context omitted.

It's clear Claude adapted code directly from the OxCaml implementation (the PR author said he pointed Claude at that code [1] and then provides a ChatGPT analysis [2] that really highlights the plagiarism, but ultimately comes to the conclusion that it isn't plagiarized). Either that highlights someone who is incompetent or they are willfully being blasé. Neither bodes well for contributing code while respecting copy…

The key is that AI adapted, not stole. It's actually capable of reasoning and generating derivative code and not just copying stuff wholesale. See examples at the bottom of my post: https://joel.id/ai-will-write-your-next-compiler/

[dead]

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

#318

Post by delian66 »

yodsanklai wrote: post #170
benterix wrote: post #62

Did these Ocaml maintainers undergo some special course for dealing with difficult people? They show enormous amounts of maturity and patience. I'd just give the offender Torvalds' treatment and block them from the repo, case closed.

In my big tech company, you don't want to be dismissive of AI if you don't want to sound like a paria. It's hard to believe how much faith leadership has in AI. They really want every engineer to use AI as much as possible. Reviewing is increasingly done by AI as well. That being said, I don't think that's why reviewers here were so cordial, but this is the tone you'd expect in the corporate world.

Please name it, so that we can know to avoid it and its products.

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

#319

Post by aiahs »

joelreymont wrote: post #298
edstarch wrote: post #292

Earlier quoted context omitted.

>I really, truly don't understand. This isn't just about manners, mores, or self-reflection. The inability or unwillingness to think about your behavior or its likely reception are stupefying. Shower thought: what does a typical conversation with an LLM look like? You ask it a question, or you give a command. The model spends some time writing a large wall of text, or performing some large amount of work, probably as…

I'll gladly discuss at length things that are near and dear to my heart. Facing random people in the public court of opinion is not one of them! Also, there's long-form writing in my blog posts, Twitter and Reddit.

Well if you wanna contribute (at least as a proxy) to OSS, you need to deal with people and make them want to deal with you. If you don't do that, no PR, regardless of how perfect it is, will ever be accepted. If you're so sure that your strategy for the future of development is correct, then prove it by building your own project, where you can fully decide which contributions are accepted, even those which are 100% ai generated. This should be easy, right? Once your project gains wide spread adoption, you can show everybody that you've been right all along. Until then, it's just empty talk.

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

#320

Post by LoganDark »

aiahs wrote: post #319
joelreymont wrote: post #298

Earlier quoted context omitted.

I'll gladly discuss at length things that are near and dear to my heart. Facing random people in the public court of opinion is not one of them! Also, there's long-form writing in my blog posts, Twitter and Reddit.

Well if you wanna contribute (at least as a proxy) to OSS, you need to deal with people and make them want to deal with you. If you don't do that, no PR, regardless of how perfect it is, will ever be accepted. If you're so sure that your strategy for the future of development is correct, then prove it by building your own project, where you can fully decide which contributions are accepted, even those which are 100%…

That's exactly their plan, it seems.
Post reply on HN