Steve Hsu: Chinese labs neglect science/math, focus on coding+agentic as DeepSeek shines on FrontierCode
teortaxesTex · x · 2026-10-29
Steve Hsu's take on the new AA v5 index: DeepSeek 1017 posts impressive, non-benchmaxxed scores on FrontierCode 2.0 (close to Fable 5.1, ahead of Argon and Kimi 3.1), but Chinese models' weak science+math performance drags them well below US frontier models. After talking to many Chinese labs, Hsu says they haven't prioritized science/math at all, focusing instead on coding+agentic capability without dedicated scientist/math teams; he predicts a reprioritization triggered by OpenAI's math breakthroughs and poor AA v5 scores. teortaxesTex replies "patience, people are working," citing domain-expert hiring posts.
More from Companies & People
- Bloomberg: Google rolls out Gemini 4 Argon amid internal employee skepticism — firstadopter · 2026-10-29
- Google names new frontier lab Mirendil among companies pretraining on latest TPUs — vjkaruna · 2026-10-29
- Factory VP: AI ships code daily for marketing teams but still can't write a funny tweet — sonofalli · 2026-10-29
- OpenAI's week: GPT-6.1 Terra 50% cheaper, enterprise ads, government sales push — AJChadha · 2026-10-29
- YC F26 startup Faraday launches end-to-end AI agents for manufacturers — ycombinator · 2026-10-29
- Company tries to sign up for new Gemini, gives up after an hour of inscrutable enterprise setup — anshulkundaje · 2026-10-29