Pinned Tweet
A Mathematician's job now is to de-slop, teach, build community and curate. There has never been a better time to be a pure mathematician.
In fact, I'm pretty sure the domain will grow if institutions can do their job well.
97
Shailesh retweeted
"The rise in sexual assaults by migrants is a price worth paying to end racism."
This is not a joke. This is a real article, published by a black-led publication company.
It suggests that a rape can be viewed as quasi-consensual because of "language barriers" and that a migrant raping you can be viewed as a "romantic gesture". Also, the sexual assault of a white woman can lead to "lasting love".
We have DEI rape now. And you're racist if you're against migrant rape!
107
609
3,011
51,273
Shailesh retweeted
My partner is a theoretical computer scientist (essentially a mathematician). She has mixed feelings about AI's impact on math research overall, but her feeling about today's release is one of overwhelming awe and excitement. Literal tears of joy. Quote: "so lucky to be alive rn"
OpenAI is downplaying this for PR reasons. If you’re a mathematician, you must feel like a nuclear bomb hit, and you’re at ground zero.
Reasoning models are two years-old. In that time they went from incapable of basic arithmetic to solving problems humans couldn’t solve for decades.
Math is only the beginning. AI will revolutionize the entirety of the human scientific endeavor. Most of us don’t appreciate what that means.
What a time to be alive!
24
34
660
26,558
We must stop respecting human intellect. Very likely that our intelligence barely scratches the surface of what's possible.
14
What luck! All these problems are in domains I'm somewhat of an expert in.
We’re releasing a broad range of new mathematical results produced by an internal frontier model.
We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results.
github.com/openai/math
25
Just couln't take it anymore and deleted my dot. It had sooo many issues that I just couldn't rely on it for anything.
Seems to me its great for chatting and shit but not for power-users.
Really wanted to like it too.
1
25
@hollllyli's dot, I will start using again if I subscribe to OAI again and the kinks are removed.
7
Shailesh retweeted
Last thing on OAI math. Here is what four models, without search, think it would mean "if it were real". Journalist friends, call a mathematician and ask them to explain. Then call someone in AI and science and ask "what are bottlenecks to this in other practical fields..."
13
52
519
56,270
it seems like all the pure math infra was just an elaborate jobs program for this in-group. these guys don't seem to be excited by truth-seeking
23
Shailesh retweeted
We are entering a new era of discovery now:
openai.com/index/sharing-ai-…
694
898
9,870
1,683,155
What's obvious looking at these results is that the model was not able to solve the problems it was presented, but was able to produce some partial or adjacent proofs which were valuable.
22
Shailesh retweeted
We're publishing a result demonstrating a substantial tightening to the results from OpenAI Problem #109 (integer multiplication).
Conditional on OpenAI's algorithmic interfaces, our parameter and network refinements improve the exponent saving in:
T(n)=O(n(\log n)^{1-\kappa})
from κ=2⁻¹⁸² to κ=5.8×10⁻³³ (between 2⁻¹⁰⁸ and 2⁻¹⁰⁷).
This represents a nearly 2⁷⁵ fold increase in the algorithm's exponent saving parameter.
We also establish a ceiling of κ<5.838×10⁻³³ for the stated network-counting family and cost inequalities. Our result exceeds 99% of that ceiling. Surpassing this ceiling would require improving the network bounds or cost analysis from the original result.
3/ Integer multiplication is probably the most shocking result. Similar to matrix multiplication it’s a “asymptotic reduction”. Also like the matrix result, it’s not practical.
But very few people would have predicted the previous barrier could be broken. Even slightly
9
11
195
101,663
The problem with this guy is that he's never internalised what is being built. He thinks he's being rational but he's just being blind.
That has lead to is him trying to 'gotcha' the models with his evals and trying to constantly find faults in them.
Being a believer is much less gay
What if the jagged frontier is mainly math + code (which you can push arbitrarily far with RLVR), and everything else starts to plateau because it is still bottlenecked by human generated data?
Model performance in non-verifiable areas has kept improving steadily, albeit much slower than for math and code. But is that steady improvement a side effect of a higher G (itself driven by RLVR), or only a function of the amount of new human data getting injected into training (which is still continually happening on a massive scale)?
A lot of things depend on the answer to this question
38
Shailesh retweeted
"convert all 722 preprints into 3blue1brown videos. don't make mistakes."
74
506
10,990
178,063