vatsachak

4 ms·

1.
by vatsachak 5d ago
They used an agent swarm. Could humanity? Of course; it was close to being solved. It might have taken five years
2.
by vatsachak 5d ago
THIS EXACTLY. The only regulation that we need right now is the model that's on tap
3.
by vatsachak 5d ago
You're right. But... Spending 132 billion tokens is definitely way more energy than human mathematicians would have thrown at the problem and probably would have solved within two years.
4.
by vatsachak 5d ago
He's expressed similar fears of being replaced in a recent talk
5.
by vatsachak 5d ago
No it's not though. Can OpenAI prove Navier-Stokes from a vacuum with no human knowledge?
6.
by vatsachak 5d ago
Why is Terry Tao even talking these arguments when all OpenAI did was a sophisticated brute force search with LLMs? Give them Navier-Stokes in a vacuum
7.
by vatsachak 6d ago
China should invest in an analog inference chip. It's a hail mary but why not.
8.
by vatsachak 7d ago
No. Starting in a sandbox given only the C specification and assembly code specification build a gcc equivalent (maintainability and performance) from the ground up.
9.
by vatsachak 7d ago
No, not that is does anything more... It's that gcc is an old and reliable piece of software built on abstractions that have stood the test of time.
10.
by vatsachak 7d ago
I said gcc, not a C compiler
11.
by vatsachak 7d ago
Okay tell me why that's not even close to gcc. You can use an LLM
12.
by vatsachak 7d ago
Math academia 2025 > Sorry, only epic problem solvers allowed here Math academia 2026 > We were more than just problem solvers I think people are overblowing this though. Wake me up when GPT-whatever writes gcc from scratch, then by t
13.
by vatsachak 7d ago
Timothy Gower
14.
by vatsachak 7d ago
Gr8 b8 m8
15.
by vatsachak 8d ago
Muh academic ivory tower
16.
by vatsachak 8d ago
That's awesome! Congratulations! I'd imagine that in three months when we all have access to communicating agent swarms this should be easier
17.
by vatsachak 9d ago
After AI agents get good enough the real bottleneck will be power generation and political systems.
18.
by vatsachak 10d ago
I think the idea would be making a frontier model that does this. One that is trained on multiple queries
19.
by vatsachak 10d ago
I agree with the caveat that it's more like a cracked junior engineer who can manage swarms of interns. Frontier Labs will probably survive off hype valuations but will serve the important purpose of discovering architectures/tech
20.
by vatsachak 11d ago
Yeah it would be quite trivial to try and implement an auto regressive AST generator for STLC with Jev provided that you had bounded variable names and integers. As you said, if it worked, they would have demoed it haha
21.
by vatsachak 11d ago
If you provide it an AST of the english language, yes.
22.
by vatsachak 11d ago
It's not an LLM though it's a frontier model on structured data
23.
by vatsachak 11d ago
Cool project! That's what I was insinuating through "better encoder"; the model creating more efficient representations of ASTs using something like JEPA
24.
by vatsachak 11d ago
It could be used for coding if you gave it an AST. If you work at TypeSafe please try this. Side note: This is probably how LLMs would perform with better encoders and next-latent prediction, so eventually those will beat this architecture
25.
by vatsachak 11d ago
An interesting example was the neomacs post on the frontpage yesterday. https://neomacs.org Some buttons on the front page don't even work. And then you have Joachim Breitner using Claude Code to produce Soundness proofs in
26.
by vatsachak 11d ago
We really are entering the AI economy. Now if only we knew if the stonks would go up or down (due to global turmoil) before I throw my savings at the SPY
27.
by vatsachak 11d ago
Given how fast computers are these days there is probably not much overhead in having emacs run as the interpreter for an incrementally compiled language with HM types
28.
by vatsachak 11d ago
I mean emacs has way more bugs than modern editors. Types would help you fix that and let you optimize code more since they help with "proving" that your code is correct. I feel like emacs could be way better if e-lisp was typed b
29.
by vatsachak 11d ago
I guess I was thinking about the NFL but you're right about college. Your point also extends the analogy, the AI might seem enough at lower levels due to that reason
30.
by vatsachak 11d ago
I disagree that AI has broken the proxies. Football analogy; AI is the wide receiver and the human is the quarterback. Even if you're the best WR in the game you're still not producing touchdowns unless you have a decent QB. It&#x

More ›