Binomial distribution: If X has the Bin(n,p) distribution for some integer n>=1 and 0<=p<=1 then P(X>m) can be written as a sum over binomial coefficients from m+1 up to n. However, it can also be written as a single integral term.

nice writeup of Claude Code under the hood medium.com/@outsightai/pe…

Yeah, AI progress is totally, definitely stalling... Look at MathArena Apex. GPT-5.1 scored 1%. Gemini 3 scored 23%. That is a >20x jump on one of the hardest reasoning tasks we have. But sure, keep your head in the sand...

Gemini 3.0 Pro is insane for coding! But is it good for biomedicine? It turns out the new update has incremental improvements in super hard questions. Supported by OpenRouter

Claude Opus 4.5 achieves a massive 80.0% on ARC-AGI-1

Have to switch model every week these days

Did Anthropic really do it? They eliminated "You are absolutely right!" :O

big if true

> The natural response to competition is to try to be different. So if you were to put multiple agents together and you tell them, “You all need to work on some problem and you are an agent and you’re inspecting what everyone else is working,” they’re going to say, “Well, if

apparently Opus 4.5 will start THINKING INSIDE OF FILES if you turn reasoning traces off

Interesting, last week ttao proved a weaker version arithmetic Kakeya conjecture that was suggested by alphaevolve (1.6 below figure) Kinda cool that not only AI systems assist with difficult theorems but also actively inspires / extends best researchers beyond 'mere instruct

Social media tends to frame AI debate into two caricatures: (A) Skeptics who think LLMs are doomed and AI is a bunch of hype. (B) Fanatics who think we have all the ingredients and superintelligence is imminent. But if you read what leading researchers actually say (beyond the

What was it again about AI finding solutions to Erdos problems? Boris Alexeev found the solution to #124 that has been open for 30 years, solution is 100% AI generated. Details here: erdosproblems.com/forum/thread/1…

DeepSeekMath-V2's use of "meta verifiers" makes me wonder if Leslie Lamport's "structured proofs" could make a comeback in the AI math scene. His proof format is a middle ground between natural language and formal proofs, where every step can be recursively expanded for more

Will this be possible in 2026?

we are at risk of running out of Erdős problems is anyone working on AI to make new ones?

This interesting week started with DeepSeek V3.2! I just wrote up a technical tour of the predecessors and components that led up to this: 🔗 magazine.sebastianraschka.com/p/technical-de… - Multi-Head Latent Attention - RLVR - Sparse Attention - Self-Verification - GRPO Updates

Here's my writeup of the proof of Telgarsky's conjecture. not independent of the AI proof. I have already seen the 'informal proof' file in the github repo. But it confirms my suspicion that it's similar to my earlier answer to a different (related) question by Aryeh Kontorovich