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

Andrej Karpathy: Software in the era of AI [video]

youtube.com

Post reply on HN
Print view View original

241–250 of 827 posts

Re: Andrej Karpathy: Software in the era of AI [video]

#241

Post by the_mitsuhiko »

yahoozoo wrote: post #211
abdullin wrote: post #102

Tight feedback loops are the key in working productively with software. I see that in codebases up to 700k lines of code (legacy 30yo 4GL ERP systems). The best part is that AI-driven systems are fine with running even more tight loops than what a sane human would tolerate. Eg. running full linting, testing and E2E/simulation suite after any minor change. Or generating 4 versions of PR for the same task so that the h…

The problem is that every time you run your full automation with linting and tests, you’re filling up the context window more and more. I don’t know how people using Claude do it with its <300k context window. I get the “your message will exceed the length of this chat” message so many times.

I started to use sub agents for that. That does not pollute the context as much

Re: Andrej Karpathy: Software in the era of AI [video]

#242

Post by diggan »

kaycey2022 wrote: post #237

I hope this excellent talk brings some much needed sense into the discourse around vibe coding.

If anything I wished the conversation turned away from "vibe-coding" which was essentially coined as a "lol look at this go" thing, but media and corporations somehow picked up as "This is the new workflow all developers are adopting".

LLMs as another tool in your toolbox? Sure, use it where it makes sense, don't try to make them do 100% of everything.

LLMs as a "English to E2E product I'm charging for"? Lets maybe make sure the thing works well as a tool before letting it be responsible for stuff.

Re: Andrej Karpathy: Software in the era of AI [video]

#243

Post by dist-epoch »

khalic wrote: post #177

His dismissal of smaller and local models suggests he underestimates their improvement potential. Give phi4 a run and see what I mean.

I tried the local small models. They are slow, much less capable, and ironically much more expensive to run than the frontier cloud models.

Re: Andrej Karpathy: Software in the era of AI [video]

#244

Post by khalic »

diggan wrote: post #191
khalic wrote: post #177

His dismissal of smaller and local models suggests he underestimates their improvement potential. Give phi4 a run and see what I mean.

> suggests a lack of understanding of these smaller models capabilities If anything, you're showing a lack of understanding of what he was talking about. The context is this specific time, where we're early in a ecosystem and things are expensive and likely centralized (ala mainframes) but if his analogy/prediction is correct, we'll have a "Linux" moment in the future where that equation changes (again) and local mod…

He understands the technical part, of course, I was referring to his prediction that large models will be always be necessary.

There is a point where an LLM is good enough for most tasks, I don’t need a megamind AI in order to greet clients, and both large and small/medium model size are getting there, with the large models hitting a computing/energy demand barrier. The small models won’t hit that barrier anytime soon.

Re: Andrej Karpathy: Software in the era of AI [video]

#245

Post by crmi »

sensanaty wrote: post #187
mentalgear wrote: post #157

Meanwhile, I asked this morning Claude 4 to write a simple EXIF normalizer. After two rounds of prompting it to double-check its code, I still had to point out that it makes no sense to load the entire image for re-orientating if the EXIF orientation is fine in the first place. Vibe vs reality, and anyone actually working in the space daily can attest how brittle these systems are. Maybe this changes in SWE with more…

There's also those instances where Microsoft unleashed Copilot on the .NET repo, and it resulted in the most hilariously terrible PRs that required the maintainers to basically tell Copilot every single step it should take to fix the issue. They were basically writing the PRs themselves at that point, except doing it through an intermediary that was much dumber, slower and less practical than them. And don't get me s…

I've got a working theory that models perform differently when used in different timezones... As in during US working hours they dont work as well due to high load. When used at 'offpeak' hours not only are they (obviously) snappier but the outputs appear to be a higher standard. Thought this for a while but now noticing with Claude4 [thinking] recently. Textbook case of anecdata of course though.

Re: Andrej Karpathy: Software in the era of AI [video]

#246

Post by khalic »

mprovost wrote: post #225
khalic wrote: post #177

His dismissal of smaller and local models suggests he underestimates their improvement potential. Give phi4 a run and see what I mean.

You can disagree with his conclusions but I don't think his understanding of small models is up for debate. This is the person who created micrograd/makemore/nanoGPT and who has produced a ton of educational materials showing how to build small and local models.

I’m going to edit, it was badly formulated, he underestimates their potential for growth is what I meant by that

Re: Andrej Karpathy: Software in the era of AI [video]

#247

Post by dist-epoch »

kubb wrote: post #174
ramon156 wrote: post #172

Earlier quoted context omitted.

The real question is how long it'll take until they're not brittle

Or will they ever be reliable. Your question is already making an assumption.

I remember when people were saying here on HN that AIs will never be able to generate picture of hands with just 5 fingers because they just "don't have common sense"

Re: Andrej Karpathy: Software in the era of AI [video]

#248

Post by hombre_fatal »

mentalgear wrote: post #157

Meanwhile, I asked this morning Claude 4 to write a simple EXIF normalizer. After two rounds of prompting it to double-check its code, I still had to point out that it makes no sense to load the entire image for re-orientating if the EXIF orientation is fine in the first place. Vibe vs reality, and anyone actually working in the space daily can attest how brittle these systems are. Maybe this changes in SWE with more…

On the other hand, posts like this are like watching someone writing ask jeeves search queries into google 20 years ago and then gesturing how google sucks while everyone else in the room has figured out how to be productive with it and cringes at his "boomer" queries.

If you're still struggling to make LLMs useful for you by now, you should probably ask someone. Don't let other noobs on HN +1'ing you hold you back.

Re: Andrej Karpathy: Software in the era of AI [video]

#249

Post by khalic »

diggan wrote: post #191
khalic wrote: post #177

His dismissal of smaller and local models suggests he underestimates their improvement potential. Give phi4 a run and see what I mean.

> suggests a lack of understanding of these smaller models capabilities If anything, you're showing a lack of understanding of what he was talking about. The context is this specific time, where we're early in a ecosystem and things are expensive and likely centralized (ala mainframes) but if his analogy/prediction is correct, we'll have a "Linux" moment in the future where that equation changes (again) and local mod…

I edited to make it clearer

Re: Andrej Karpathy: Software in the era of AI [video]

#250

Post by TeMPOraL »

diggan wrote: post #230
TeMPOraL wrote: post #223

Earlier quoted context omitted.

That's literally the wrong way to use LLMs though. LLMs think in tokens, the less they emit the dumber they are, so asking them to be concise, or to give the answer before explanation, is extremely counterproductive.

I was trying to make a point regarding "reliability", not a point about how to prompt or how to use them for work.

This is relevant. Your example may be simple enough, but for anything more complex, letting the model have its space to think/compute is critical to reliability - if you starve it for compute, you'll get more errors/hallucinations.
Post reply on HN