Wait. What. 1GW IDC for Zhipu alone?! This is ahead of my timeline. This is safely enough to build a "mythos" assuming some workable (even H800-level)…
RT Larry DialQuestion for the LLM Research Community: Is anyone aware of fully reproducible experimental results showing that MLA beats GQA under the same KV cache?…
worth reading. That's for once Dean actually speaking-as-OpenAIDean W. Ball: As the functional time horizon of frontier AI systems grows longer, novel risks can emerge. Today,…
feel the acceleration anon:in 2026, 3-4 days is what 2 more weeks used to be in 2025…RedRibbon: @teortaxesTex omfg have faith bro. We're 3-4 days away…
RT Philippe LemoineI think it's now clear that Dean Ball's post wasn't random but was a public manifestation of a debate that is currently taking place…
RT Daniel LittThis is actually incredible/incredibly funny.Andy Jiang: @littmath @__alpoge__ GPT:Take π: P¹ × Sym²(P¹) → Sym³(P¹), (p, {q,r}) ↦ {p,q,r}. R be its ramification divisor;H…
> It was running E2E tests, validating desktop, tablet, and mobile, taking screenshots, finding failures, fixing them, even adapting for low end hardware and low FPS…
Though, that said…Has he ever left? Using Dean to preannounce the measure as "analysis" and "prediction" and take the heat off the labs would be quite…
Will it work?Josh: 5 hour update: some guy named after a welsh castle is gonna give me a thousand dollars pretty soon. still dont really understand…
> Be Dr. Chris Fall> take the post of the director of the U.S. Center for AI Standards and Innovation (CAISI) under NIST> American models are…