I'm not an expert on the subject itself, but I can tell that the transcript, in its entirety, is missing a solid line. While the parts of this talk are great on their own, I feel they couldn't stitch the whole story together well. And probably he might not be confident of completeness and composition of his thought. What's the whole point? That should be answered in the first few minutes.
Andrej Karpathy's talk on the future of the industry
111–120 of 193 posts
Re: Andrej Karpathy's talk on the future of the industry
#112Earlier quoted context omitted.
Do you have anything to share on that workflow? I've been trying to get a local-first AI thing going, could use your insights!
It's super easy. I already had llama.cpp/llama-server set up for a bunch of other stuff and actually had my own homebrew RAG dialog engine, aider is just way better. One crazy thing is that since I keep all my PIM data in git in flat text I now have essentially "siri for Linux" too if I want it. It's a great example of what Karpathy was talking about where improvements in the ML model have consumed the older decision…
I really appreciate you going into such technical specificity, thank you! I'll have to steal that siri for linux setup, that sounds awesome. Exploring ways to make use of compute people have lying around to do useful things without the vendor dependencies. But I'm relatively new to the AI scene, so your input really boosts my learning speed, thank you again!
Re: Andrej Karpathy's talk on the future of the industry
#113Christ I'm gonna be forced to listen to the moronic managers and C-suites repeat this "software 3.0" bullshit incessantly from now on aren't I...
Re: Andrej Karpathy's talk on the future of the industry
#114Earlier quoted context omitted.
So... Who builds the AI? This is why I think the AI industry is mostly smoke and mirrors. If these tools are really as revolutionary as they claim they are, then they should be able to build better versions of themselves, and we should be seeing exponential improvements of their capabilities. Yet in the last year or so we've seen marginal improvements based mainly on increasing the scale and quality of the data they'…
> If these tools are really as revolutionary as they claim they are, then they should be able to build better versions of themselves, and we should be seeing exponential improvements of their capabilities. Recursive self-improvement is literally the endgame scenario - hard takeoff, singularity, the works. Are you really saying you're dissatisfied with the progress of those tools because they didn't manage to end the…
Moving goal posts? This was a response to the claim that AIs are the new code.
Re: Andrej Karpathy's talk on the future of the industry
#115If you want to try what Karpathy is describing live today, here's a demo I wrote a few months ago: https://universal.oroborus.org/
It takes mouse clicks, sends them to the LLM, and asks it to render static HTML+CSS of the output frame. HTML+CSS is basically a JPEG here, the original implementation WAS JPEG but diffusion models can't do accurate enough text yet.
My conclusions from doing this project and interacting with the result were: if LLMs keep scaling in performance and cost, programming languages are going to fade away. The long-term future won't be LLMs writing code, it'll be LLMs doing direct computation.
Re: Andrej Karpathy's talk on the future of the industry
#116I might be wrong, but it seems like some people are misinterpreting what is being said here. Software 3.0 isn't about using AI to write code. It's about using AI instead of code. So not Human -> AI -> Create Code -> Compile Code -> Code Runs -> The Magic Happens. Instead, it's Human -> AI -> The Magic Happens.
That jibes with what Nadella said in an interview not too long ago. Essentially, SaaS apps disappear entirely as LLMs interface directly with the underlying data store. The unspoken implication being that software as we understand it goes away as people interface with LLMs directly rather than ~~computers~~ software at all.
I kind of expect that from someone heading a company that appears to have sold-the-farm in an AI gamble. It’s interesting to see a similar viewpoint here (all biases considered)
Re: Andrej Karpathy's talk on the future of the industry
#117Earlier quoted context omitted.
Objectivity is lacking throughout the entire talk, not only in the thesis. But objectivity isn't very good for building hype.
I'm curious why you think that? I thought the talk was pretty grounded. There was a lot of skepticism of using LLMs unbounded to write software and an insistence on using ground truth free from LLM hallucination. The main thesis, to me, seemed like "we need to write software that was designed with human-centric APIs and UI patterns to now use an LLM layer in front and that'll be a lot of opportunity for software engi…
It's a lot of meandering and mundane analogies that don't work very well or explain much, so it's totally understandable that so many people have different interpretations of what he's even trying to say. The only consistent takeaway here is that he's talking about using AI (of many sorts) alongside legacy software.
Re: Andrej Karpathy's talk on the future of the industry
#118For context - I was in the audience when Karpathy gave this amazing talk on software 3.0. YC has said the official video will take a few weeks to release, by which Karpathy himself said the talk will be deprecated. https://x.com/karpathy/status/1935077692258558443
What a poor judgement he must have if his outlook becomes irrelevant in a few weeks' time. Edit: the emoji at the end of the original sentence has not been quoted. How a smile makes the difference. Original tweet: https://x.com/karpathy/status/1935077692258558443
the way I read it it's more about how fast examples and references become irrelevant. fundamentals of the speech not.
Re: Andrej Karpathy's talk on the future of the industry
#119Earlier quoted context omitted.
> If these tools are really as revolutionary as they claim they are, then they should be able to build better versions of themselves, and we should be seeing exponential improvements of their capabilities. Recursive self-improvement is literally the endgame scenario - hard takeoff, singularity, the works. Are you really saying you're dissatisfied with the progress of those tools because they didn't manage to end the…
Nope. Just that 1. Is better than people. 2. Isn't better than people. Pick one! If the former then yes singularity. The only hope is it's "good will" (wouldn't bet on that) or turning off switches. If the latter you still need more workers (programmers or whatever they'll be called) due to increased demand for compute solutions.
> Nope. Just that 1. Is better than people. 2. Isn't better than people. Pick one!
That's too coarse of a choice. It's better than people at increasingly large number of distinct tasks. But it's not good enough to recursively self-improve just yet - though it is doing it indirectly: it's useful enough to aid researchers and businesses in creating next generation of models. So in a way, the recursion and resulting exponent are already there, we're just in such early stages that it looks like linear progress.
Re: Andrej Karpathy's talk on the future of the industry
#120Earlier quoted context omitted.
so you took my transcript and put it behind a newsletter sub? haha. just share them!
not quite, i compiled the slides within a few hours of the talk yesterday well before your transcript was available. the slides are my main output/contribution. a full slides+transcript is too long for substack. i've linked your transcript prominently for people to find, and used it to fix slide ordering because twitter people took terrible notes for the purpose of exact talk reconstruction. i exepct YC to prioritize…
Looks like you are putting a derivative behind a paywall though, no? I think quid pro quo let pudiklubi publish your work too? Some kind of open license?