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

A study on robustness and reliability of large language model code generation

arxiv.org

Post reply on HN
Print view View original

151–160 of 229 posts

Re: A study on robustness and reliability of large language model code generation

#151

Post by FeepingCreature »

Gigachad wrote: post #138
capableweb wrote: post #25

Earlier quoted context omitted.

Which version of GPT are you using? I've found GPT4 moderately helpful, like getting help from someone with a wide but shallow experience of lots of things. > I can't easily fact check it Why not? How did you find out things about basic terminology or concepts you're not familiar about before GPT? Apply the same methods, although you just have to fact check rather than coming up with what to check, so removes a small…

>Why not? How did you find out things about basic terminology or concepts you're not familiar about before GPT? Because if it tells me something that could just be found on google, I could have found it there first and not have to do as much fact checking. And if it tells me something I can't easily find, I can't tell if it's got some deep insight on harder to locate info, or if it's just made up.

ChatGPT gives you ideas and terms that you can google. That's often a rarity in a new language/framework.

Re: A study on robustness and reliability of large language model code generation

#152

Post by FeepingCreature »

Gigachad wrote: post #14

I feel like AI for programming has been so overhyped. I've attempted to use ChatGPT for programming so many times, and almost every time it's just wasted my time. Giving me outright lies or generating stuff that looks right but doesn't work. And just throwing out it's answer and starting from scratch is faster than fixing its output. I've only found it useful for explaining basic terminology or concepts for a topic I…

My experience is that ChatGPT often makes errors, but the errors are different than the errors I'd make. So it's easy to work with it collaboratively: I take a turn, GPT takes a turn, I take a turn... and it has an extremely broad spectrum of surface-level knowledge about tech.

The most annoying part is it's impossible for it to say that it doesn't know something. Hard to train for, I know.

Re: A study on robustness and reliability of large language model code generation

#153

Post by FeepingCreature »

jsunderland323 wrote: post #115
corethree wrote: post #107

Earlier quoted context omitted.

>You’re implying 100% reliability is actually attainable. If that were the case, wouldn’t that mean the halting problem would have been solved by AI? I’m not an expert but I’ve heard that’s like one of those fundamental laws of information theory that really can’t be broken. Nah even 100% reliability isn't attainable by a human. 100% obviously doesn't imply solving the halting problem. 100% as in 100% as reliable as…

I'm not going to rebut everything but I want to point out that I think Geoffrey Hinton is way more in your camp in regards to faith in the technology (which is what we're discussing). I'm in the camp that thinks LLMs are at best a viable competitor to furby -- I have zero existential fears and have very little confidence in them. I'm not opposed to AIs, I just think the state of the art is really really bad (sorry) a…

Your hypothesis is "our company will destroy the world" is a marketing move?

Look. People in reality are basically never this gigabrained. Maybe consider as your first-line theory, that when people say "AI will destroy the world", that they mean to express that they believe that AI will destroy the world?

Re: A study on robustness and reliability of large language model code generation

#154

Post by FeepingCreature »

rvz wrote: post #54

I can guarantee you that no-one here fully trusts any LLM to write complicated bug free code without you (a human) to keep checking over it. Anyone with an understanding of unexplainable black-box AIs knows that LLMs hallucinating is not 'the same thing as a human' . Humans can be held accountable for their mistakes and can explain themselves transparently. LLMs fundamentally cannot reason or explain themselves trans…

Of course, humans also often try to bullshit their way through wrong answers.

Re: A study on robustness and reliability of large language model code generation

#155

Post by maxbond »

robwwilliams wrote: post #150

Is this worth reading? When authors cannot write a grammatically correct abstract I tune out. Try to make sense of this monster sentence: “The misuse of APIs in the generated code could lead to severe problem, such as resource leaks, program crashes, this http URL make things worse, the users of LLM code generation services are actually the developers that are most vulnerable to these code that seems right -- They ar…

Bear in mind that most researchers are not native English speakers.

Re: A study on robustness and reliability of large language model code generation

#156

Post by Tiberium »

maxbond wrote: post #155
robwwilliams wrote: post #150

Is this worth reading? When authors cannot write a grammatically correct abstract I tune out. Try to make sense of this monster sentence: “The misuse of APIs in the generated code could lead to severe problem, such as resource leaks, program crashes, this http URL make things worse, the users of LLM code generation services are actually the developers that are most vulnerable to these code that seems right -- They ar…

Bear in mind that most researchers are not native English speakers.

This argument is not really valid anymore, especially in a paper that talks about LLMs.

Re: A study on robustness and reliability of large language model code generation

#157

Post by maxbond »

Tiberium wrote: post #156
maxbond wrote: post #155

Earlier quoted context omitted.

Bear in mind that most researchers are not native English speakers.

This argument is not really valid anymore, especially in a paper that talks about LLMs.

It's an observation, not an argument, but note that this is a paper about LLM mistakes. But if the language bothers you and you view LLMs as the solution, you're certainly free to feed it into an LLM yourself.

To be frank, I find it a little off-putting to suggest being a non-native English speaker is "no longer valid."

Re: A study on robustness and reliability of large language model code generation

#158

Post by robwwilliams »

jprete wrote: post #17
willsmith72 wrote: post #13

It's an interesting idea, but why is this written so weirdly? "the users of LLM code generation services are actually the developers that are most vulnerable to these code that seems right -- They are always novice developers that are not familiar with the APIs that LLMs generate code for them" It looks like they could've benefited from an LLM doing a pass-over. Are they saying it's only "novice developers" using LLM…

I think the author is simply weak at English, although there are some typos that really should have been corrected even in non-idiomatic English (e.g. missing spaces between words). Really, as long as the premise, evidence, and results are clear and reproducible, the quality of the English doesn’t much matter.

Writing is code too. Writing English well does matter if you want to have your work read.

Re: A study on robustness and reliability of large language model code generation

#159

Post by boxed »

maxbond wrote: post #157
Tiberium wrote: post #156

Earlier quoted context omitted.

This argument is not really valid anymore, especially in a paper that talks about LLMs.

It's an observation, not an argument, but note that this is a paper about LLM mistakes . But if the language bothers you and you view LLMs as the solution, you're certainly free to feed it into an LLM yourself. To be frank, I find it a little off-putting to suggest being a non-native English speaker is "no longer valid."

Maybe the previous commenter meant that they could have used an LLM to translate from Mandarin to English to get a better sentence?

Which is... ironic I think :P

Re: A study on robustness and reliability of large language model code generation

#160

Post by barbarr »

Tiberium wrote: post #156
maxbond wrote: post #155

Earlier quoted context omitted.

Bear in mind that most researchers are not native English speakers.

This argument is not really valid anymore, especially in a paper that talks about LLMs.

I found it somewhat charming that the authors, who express skepticism about the use of LLMs by novices, indeed stuck to their views and decided not to run their phrasings through an LLM. This will probably be one of the dwindling number of papers released each year with "fairly bad" grammar - and part of me wonders if we're losing something in the process of LLM-meditated homogenization.
Post reply on HN