
At first slowly, and then all at once.
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
sharan retweeted
This week's episode on Subtext is with @ponnappa, founder of @realfastai.
Many on X (and Bangalore tech) don't need an introduction to him. Those who do should know he's one of the smartest founders thinking about what an AI-native firm looks like.
Full episode out tomorrow!
agent swarms are beautiful! I love how token efficient bot-bot messages are compared to bot-human messages.
here is my variant as a minimal pi extension: github.com/sharanry/pi-swarm…
from the horse’s mouth. future of math from the perspective of a fields medalist. great watch!
youtu.be/6uIJdXmB4vE?is=8dNB…
open source models need a come back more than ever
We congratulate Levent Alpöge and Tristan Buckmaster on their remarkable mathematical work.
We (the researchers and the agents) did not see any of their work through any means until they released it publicly — in particular, no specific user data was accessed in order to solve this problem.
While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models.
However, our proofs differ significantly and even the precise results proved are different in the Euler case (forced vs. unforced).
one day the swarm shall take over
after reading the first huggingface writeup, we set up Swarmbook for all of our internal agents, an internal agent facing message board supporting threads and replies
now, before any agent gets started, they quickly pick up any context across the team from any other agent across our stack (coding, incident response, etc), and on conclusion / learnings, append their learnings to the relevant board / thread, replying to other agents as they go
picrel is color-coded by the self-declared agent name ( eg chunk-error-rooter, sdk-address-hardcut ) with solid lines being threads and dotted lines being replies
new workflow which I am using to deal with dense/long codex output:
- make your agent root an issue/plan a feature impl etc the usual way.
- use `eli5` command to ingest what codex is trying to say one thing at a time and build a world model.
- what is the fix?
- why did X happen?
- etc
- execute once I am convinced.
need more evals lol
Replying to @OpenAI
GPT-6 Astra is state-of-the-art on FrontierMath Tier 4, ARC-AGI 3, and TerminalBench-4.0.
GPT‑6 Astra is also a major advance for scientific discovery, with state-of-the-art performance on Terminal-Bench Science 0.1 and HealthBench Pro.
sharan retweeted
this is the right approach to open source maintenance
blog.rust-lang.org/inside-ru…











