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

An Honest Review of AI Programming

mropert.github.io

Post reply on HN
Print view View original

41–50 of 90 posts

Re: An Honest Review of AI Programming

#41

Post by kamranjon »

hahahaa wrote: post #25
Sharlin wrote: post #8

Earlier quoted context omitted.

I'd like to hear actual rebuttals to those quotes.

> Hallucinations are an inherent property of how LLMs work. While technically true the hallucination rates on modern models is low and other checks can ensure that by the time a human sees it it is most likely solid. For research there is more danger as there is less feedback loop other than other LLM scrutinising the first. For research I get it to come to a conclusion but provide me with links so I can judge. More…

“While technically true the hallucination rates on modern models is low…”

Isn’t this entirely context dependent? Where did you get the information that modern models have low hallucination rates? I’d love to see the benchmark if there is one, it seems like it would be useful to track.

Re: An Honest Review of AI Programming

#42

Post by Lwerewolf »

duskdozer wrote: post #11
lnenad wrote: post #2

> An Honest Review > I do admit that this approach immediately triggered my contrarian side and made me very defiant of any AI tool. Makes sense. > While this could be partially remedied by always asking for a primary source or citation, I dislike the idea that one has to add magical incantations to their queries to get the right results. It’s a good laugh to make fun of “make no mistake” memes, until you start havin…

It just seems silly to need to beg in English for things like that. What's the entire space of things you need to think about? Can you be sure the specific way you asked would be correct, or do you also have to add "don't lie about following this instruction:"? Why not have an interface `ai --make-no-mistakes --dont-make-things-up`? I don't want to have to talk to the computer like it's people.

You don't have to. Chatgpt-sol-high already does that for me. Could be the extra instructions or the base system prompt or whatever, point is - it already does it.

Re: An Honest Review of AI Programming

#43

Post by cyanclouds »

The case study of the LLM getting the basic Mono framework usage wrong seems outdated.

They're pretty flawless at any framework now, React, SwiftUI, Unity - literal skill issue if you can't get good code out of an LLM.

If an LLM can write a nuanced paragraph about any topic, it can certainly write a simple [insert popular framework] component which has considerably less potential variation.

Consider that there are 100k+ words in English, 90k+ in Spanish, and 77 in C# (38 in JavaScript!).

The LLM can write code.

Can it make software? No. Because software is a lot more than code.

But the LLM can write code.

Any anti-AI takes moving forward are going to have to acknowledge that I think

Re: An Honest Review of AI Programming

#44

Post by somenameforme »

cronin101 wrote: post #32

At the risk of sharing an unpopular take, this reads a lot like someone who decided how they feel about LLM-driven engineering ~5 months ago and doesn’t seem to reflect the current reality. (Or they’re working in an organization with lower budgets and not cranking the frontier models of today) I fully agree about the cost/sustainability parts, but to suggest you can’t build a high quality coding/verifying/iterating l…

Moreso it seems like nonsense that's masked by an excellent facade of eloquence. Ironic given it's a a post ranting against LLMs.

I stopped reading when he decided to go off on some tangent about how LLMs can't do native programming or games because they weren't trained on it. Ignoring the technical issues there, an even more overt one is that, amongst a zillion other projects, even the entire source code for Unreal Engine is available and within their training corpus. Even older models were quite competent at working with Unreal and outputting idiomatic code, which is saying something if anybody's ever worked with Unreal.

I increasingly think people are writing dumb articles on purpose because it drives 'engagement' more than a straight forward and accurate post would, at least on average.

Re: An Honest Review of AI Programming

#46

Post by lnenad »

righthand wrote: post #36
lnenad wrote: post #2

> An Honest Review > I do admit that this approach immediately triggered my contrarian side and made me very defiant of any AI tool. Makes sense. > While this could be partially remedied by always asking for a primary source or citation, I dislike the idea that one has to add magical incantations to their queries to get the right results. It’s a good laugh to make fun of “make no mistake” memes, until you start havin…

Not providing a source is how Llms get caught in sycophancy and hallucinations. The “tool” needs improvement, especially one that claims it’s intelligence. How would you react if you found out after schooling everything you learned has half-hallucinated because your teacher made it up and there is no text book or historical log of facts? You had assumed this whole time you were learning facts…now what?

Again the analogy isn't complete. If I went in knowing the teacher isn't an expert in the field and they might get some things wrong I'd definitely double check stuff. I'm sure you've had similar teachers in your life and you did this either way (I know I have).

Re: An Honest Review of AI Programming

#47

Post by unknownfuture »

CuriouslyC wrote: post #29
joshribakoff wrote: post #19

Earlier quoted context omitted.

Some people have the agent hallucinating its own meaningless tests

Hence the advice to read the tests? Interesting how people make the argument that AI code is garbage and can't be trusted (it's inelegant and poorly structured, sure, but that's a different matter), and how the only option isn't to just spot check the work, but to laboriously write it by hand.

> Hence the advice to read the tests?

Reading tests and actually catching issues requires a heck of a lot of effort. IME code reviews of tests are often more laborious than reviewing the code itself since so much if it is reasoning about corner cases.

I'm not saying don't do it, but the LOE is high. Done well I'd argue it runs close to the effort involved in just authoring those tests by hand.

And that's ignoring that the "vibe all the things" crowd is explicitly telling people not to read the code. At all.

(And if you think I'm exaggerating: an exec where I work decided to own rebuilding one of our services. They told me how recently the LLM generated multiple thousands of lines of code and they pushed it with minimal review, figuring we'll just fix the bugs as they happen...).

Re: An Honest Review of AI Programming

#48

Post by tangotaylor »

> In a past life I had to argue every year to renew a license for a profiling tool that cost about 20 EUR a month. I’ve heard since that everyone at the company is now getting a Claude subscription, even non-programmers.

Underrated quote. I also found this frustrating. At one company I was at (a very old company which was trying to pivot to software engineering), we had hellish bureaucratic fights with the IT department to get access to Pycharm, Obsidian, and even GitHub. But then the AI craze dropped and management just gave us all GitHub Copilot access without us even asking.

Cory Doctorow's book The Reverse Centaur's Guide to Life After AI talks more about this. It's just a modern symptom of an age-old power struggle. The workers want more control over their craft, including quality standards and tools, but their bosses want more control over the workers.

What's happening now is bosses are feeling pressure from investors to show productivity gains from using AI, so bosses panic-push AI within their companies. Which leads to misaligned incentives like tokenmaxxing.

Re: An Honest Review of AI Programming

#49

Post by cronin101 »

ryanackley wrote: post #34
cronin101 wrote: post #32

At the risk of sharing an unpopular take, this reads a lot like someone who decided how they feel about LLM-driven engineering ~5 months ago and doesn’t seem to reflect the current reality. (Or they’re working in an organization with lower budgets and not cranking the frontier models of today) I fully agree about the cost/sustainability parts, but to suggest you can’t build a high quality coding/verifying/iterating l…

His experience is completely plausible. He’s in a niche that requires highly performant code and most complex, highly performant games do nit have source available for models to train on. It’s a very common observation that the farther you stray from mainstream, the less effective the LLM models become.

Have you actually tried performance optimisation using an agent? With any programming language/framework that has quality profiling tooling (which is a prerequisite for most projects) I have had huge success with automated hotspot profiling where the LLM can propose theories, test the impact of fixes, convince you of which to pursue, etc.

High performance algorithms are quite well documented so it isn't unreasonable to expect an LLM to apply them appropriately when given the ability to "see" where they need to be applied.

Post reply on HN