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

341–350 of 827 posts

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

#341

Post by chamomeal »

eddd-ddde wrote: post #319
bandoti wrote: post #285

Earlier quoted context omitted.

Here’s a few problems I foresee: 1. People get lazy when presented with four choices they had no hand in creating, and they don’t look over the four and just click one, ignoring the others. Why? Because they have ten more of these on the go at once, diminishing their overall focus. 2. Automated tests, end-to-end sim., linting, etc—tools already exist and work at scale. They should be robust and THOROUGHLY reviewed by…

With lazy people the same applies for everything, code they do write, or code they review from peers. The issue is not the tooling, but the hands.

I am not a lazy worker but I guarantee you I will not thoroughly read through and review four PRs for the same thing

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

#342

Post by throwdbaaway »

diggan wrote: post #307
kubb wrote: post #298

Earlier quoted context omitted.

Sometimes I get "Hi!", sometimes "Hey!".

Which model? Just tried a bunch of ChatGPT, OpenAI's API, Claude, Anthropic's API and DeepSeek's API with both chat and reasonee, every single one replied with a single "hi".

o3-mini-2025-01-31 with high reasoning effort replied with "Hi" after 448 reasoning tokens.

gpt-4.5-preview-2025-02-27 replied with "Hi!"

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

#343

Post by RHSeeger »

deadbabe wrote: post #290
koakuma-chan wrote: post #278

Earlier quoted context omitted.

> That sounds awful. A truly terrible and demotivating way to work and produce anything of real quality This is the right way to work with generative AI, and it already is an extremely common and established practice when working with image generation.

It is not. The right way to work with generative AI is to get the right answer in the first shot. But it's the AI that is not living up to this promise. Reviewing 4 different versions of AI code is grossly unproductive. A human co-worker can submit one version of code and usually have it accepted with a single review, no other "versions" to verify. 4 versions means you're reading 75% more code than is necessary. Mult…

That's not really comparing apples to apples though.

> A human co-worker can submit one version of code and usually have it accepted with a single review, no other "versions" to verify.

But that human co-worker spent a lot of time generating what is being reviewed. You're trading "time saved coding" for "more time reviewing". You can't complain about the added time reviewing and then ignore all the time saved coding. THat's not to say it's necessarily a win, but it _is_ a tradeoff.

Plus that co-worker may very well have spent some time discussing various approaches to the problem (with you), with is somewhat parallel to the idea of reviewing 4 different PRs.

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

#344

Post by SirMaster »

elif wrote: post #308
osigurdson wrote: post #296

Earlier quoted context omitted.

I'm not sure that AI code has to be sloppy. I've had some success with hand coding some examples and then asking codex to rigorously adhere to prior conventions. This can end up with very self consistent code. Agree though on the "pick the best PR" workflow. This is pure model training work and you should be compensated for it.

Yep this is what Andrej talks about around 20 minutes into this talk. You have to be extremely verbose in describing all of your requirements. There is seemingly no such thing as too much detail. The second you start being vague, even if it WOULD be clear to a person with common sense, the LLM views that vagueness as a potential aspect of it's own creative liberty.

I'm really waiting for AI to get on par with the common sense of most humans in their respective fields.

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

#345

Post by sensanaty »

kypro wrote: post #229
sensanaty wrote: post #187

Earlier quoted context omitted.

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 think part of the problem is that code quality is somewhat subjective and developers are of different skill levels. If you're fine with things that kinda working okay and you're not the best developer yourself then you probably think coding agents work really really well because the slop they produce isn't that much worse than yourself. In fact I know a mid-level dev who believes agent AIs write better code than hi…

Listen, I won't pretend to be The God Emperor Of Writing Code or anything of the sort, I'm realistically quite mediocre/dead average in the grand scheme of things.

But literally yesterday, with Claude Code running 4 opus (aka: The latest and greatest, to intercept the "dId YoU tRy X" comment) which has full access to my entire Vue codebase at work, that has dedicated rules files I pass to it, that can see the fucking `.vue` file extension on every file in the codebase, after prompting it to "generate this vue component that does X, Y and Z" spat out React code at me.

You don't have to be Bjarne Stroustrup to get annoyed at this kinda stuff, and it happens constantly for a billion tiny things on the daily. The biggest pushers of AI have finally started admitting that it's not literally perfect, but am I really supposed to pretend that this workflow of having AIs generate dozens of PRs where a single one is somewhat acceptable is somehow efficient or good?

It's great for random one-offs, sure, but is that really deserving of this much insane, blind hype?

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

#346

Post by 9rx »

jonny_eh wrote: post #82
hgl wrote: post #51

It’s fascinating to think about what true GUI for LLM could be like. It immediately makes me think a LLM that can generate a customized GUI for the topic at hand where you can interact with in a non-linear way.

An ever-shifting UI sounds unlearnable, and therefore unusable.

Tools like v0 are a primitive example of what the above is talking about. The UI maintains familiar conventions, but is laid out dynamically based on surrounding context. I'm sure there are still weird edge cases, but for the most part people have no trouble figuring out how to use the output of such tools already.

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

#347

Post by eitally »

It's going to be very interesting to see how things evolve in enterprise IT, especially but not exclusively in regulated industries. As more SaaS services are at least partly vibe coded, how are CIOs going to understand and mitigate risk? As more internal developers are using LLM-powered coding interfaces and become less clear on exactly how their resulting code works, how will that codebase be maintained and incrementally updated with new features, especially in solo dev teams (which is common)?

I easily see a huge future for agentic assistance in the enterprise, but I struggle mightily to see how many IT leaders would accept the output code of something like a menugen app as production-viable.

Additionally, if you're licensing code from external vendors who've built their own products at least partly through LLM-driven superpowers, how do you have faith that they know how things work and won't inadvertently break something they don't know how to fix? This goes for niche tools (like Clerk, or Polar.sh or similar) as much as for big heavy things (like a CRM or ERP).

I was on the CEO track about ten years ago and left it for a new career in big tech, and I don't envy the folks currently trying to figure out the future of safe, secure IT in the enterprise.

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

#348

Post by bawana »

infecto wrote: post #281
darqis wrote: post #193

when I started coding at the age of 11 in machine code and assembly on the C64, the dream was to create software that creates software. Nowadays it's almost reality, almost because the devil is always in the details. When you're used to write code, writing code is relatively fast. You need this knowledge to debug issues with generated code. However you're now telling AI to fix the bugs in the generated code. I see it…

This is most definitely not toys for rich people. Now perhaps depending on your country it may be considered rich but I would comfortably say that for most of the developed world, the costs for these tools are absolutely attainable, there is a reason ChatGPT has such a large subscriber base. Also the disconnect for me here is I think back on the cost of electronics, prices for the level of compute have generally gone…

I suspect that economies of scale are different for software and hardware. With hardware, iteration results in optimization of the supply chain, volume discount as the marginal cost is so much less than the fixed cost, and lower prices in time. The purpose of the device remains fixed. With software, the software becomes ever more complex with technical debt - featuritis, patches, bugs, vulnerabilities, and evolution of purpose to try and capture more disparate functions under one environment in an attempt to capture and lock in users. Price tends to increase in time. (This trajectory incidentally is the opposite of the unix philosophy - having multiple small fast independent tools than can be concatenated to achieve a purpose.) This results in ever increasing profits for software and decreasing profits for hardware at equilibrium. In the development of AI we are already seeing this-first we had gpt, then chatbots, then agents, now integration with existing software architectures.Not only is each model ever larger and more complex (RNN->transformer->multihead-> add fine tuning/LoRA-> add MCP), but the bean counters will find ways to make you pay for each added feature. And bugs will multiply. Already prompt injection attacks are a concern so now another layer is needed to mitigate those.

For the general public, these increasing costs will besubsidized by advertising. I cant wait for ads to start appearring in chatGPT- it will be very insidious as the advertising will be comingled with the output so there will be no way to avoid it.

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

#349

Post by NitpickLawyer »

guappa wrote: post #301
infecto wrote: post #283

Earlier quoted context omitted.

LLM APIs are pretty darn cheap for most of the developed worlds income levels.

Yeah, because they're bleeding money like crazy now. You should consider how much it actually costs, not how much they charge. How do people fail to consider this?

No, there are 3rd party providers that run open-weights models and they are (most likely) not bleeding money. Their prices are kind of similar, and make sense in a napkin-math kind of way (we looked into this when ordering hardware).

You are correct that some providers might reduce prices for market capture, but the alternatives are still cheap, and some are close to being competitive in quality to the API providers.

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

#350

Post by diggan »

SirMaster wrote: post #344
elif wrote: post #308

Earlier quoted context omitted.

Yep this is what Andrej talks about around 20 minutes into this talk. You have to be extremely verbose in describing all of your requirements. There is seemingly no such thing as too much detail. The second you start being vague, even if it WOULD be clear to a person with common sense, the LLM views that vagueness as a potential aspect of it's own creative liberty.

I'm really waiting for AI to get on par with the common sense of most humans in their respective fields.

I think you'll be waiting for a very long time. Right now we have programmable LLMs, so if you're not getting the results, you need to reprogram it to give the results you want.
Post reply on HN