RT Elliot ArledgeFor those wondering why I use a Kimi Linear megakernel instead of Qwen 3.6, first look at the parameter counts. One is 35 billion,…
Maturity.Elliot Arledge: Claude Fable 5 [max] wrote the first genuine (and fastest) megakernel ever submitted to KernelBench-Mega.It was tested on: Kimi-Linear W4A16 batch-1 decode for RTX…
We need to convince EAsians that nootropics work at boosting scores on their shitty life-breaking evals^W exams like gaokao. Surely they'll soon hill-climb to something better…
I remember @ConradBastable teaching me to respect financialism. It is indeed one of the absurd, broken magicks of our civilization – like math, like faith. But…
The art of Spiritual Victory.This is Elite Sexpat (sub)Human CapitalThis is the purebred warrior of Hajnal!…He has no clue what the fuck Chinese datacenters are powered…
RT davincipre-chatgpt openai was a lab. pre-gemini deepmind as well (still somewhat is, maybe?). anthropic almost never was (it's an extremely product-oriented company with very little…
I'd also say we should have more technical reports (if not weights). The "safety" rationale for only sharing Safety Evals is shallow, fellow GPU-rich labs have…
So apparently Gemini Omni Flash is to Seedance 2.0 what Seedance 2.0 is to Veo 3. But Seedance 2.0 curb stomped Veo 3… is this real?…
RT Kardashev Age VikingRe @teortaxesTex Fixed.
V4 is really bad at in-context learning, alasMerely GLM 5.0 levelwe shall see how much they can improveDeyao Zhu: [6/n] We evaluated model releases from September…