AI Daily: Tencent Hunyuan defends smaller models, Lean certifies Claude's Riemann proof
创业邦 · wechat · 2026-11-22
Tencent's chief AI scientist Yao Shunyu and the Hunyuan team hosted an overseas AMA, explaining why they stopped at 700B parameters — Hy3/Hy4 proved smaller models can beat larger ones at lower cost — while noting Hy4 still has big headroom via post-training, with heavy investment in large-scale RL and plans for larger models. Separately, LeanFRO and Mathlib Initiative officially certified that Levent Alpöge's Riemann hypothesis proof built with Claude is correct, verifying 17M lines of Lean code in a single day.
Other highlights: Nestlé partners with Microsoft, Boston Dynamics and Google on agentic AI factory lines; OpenAI released MedChemBench (99 questions, GPT-6.2 Astra scoring 84.5%); stealth anonymous model Gestalt topped Contra's Frontier Design Benchmark with Elo 1931, beating human reference works; Musk said FSD V15 rolls out next month; Databricks is in early talks for funding at a $225B valuation; SoftBank reportedly once explored a $1T acquisition of SK Hynix.
More from Companies & People
- Taxi Driver screenwriter Paul Schrader, 80, embraces AI: 'opening the cage of a monster' — nordicinst · 2026-11-22
- Google DeepMind and YC startup Hebbian host world's first Robot Use Agents hackathon — DynamicWebPaige · 2026-11-22
- Obsidian CEO: Top designers wasted prime years on big tech minutiae, AI reawakens their talent — Jackyhuang · 2026-11-22
- Howard Marks: AI's pace of progress makes OpenAI and Anthropic hard to value — WhatTheLJW · 2026-11-22
- OpenAI ships MedChemBench; Tencent Hunyun team confirms smaller models on roadmap — 创业邦 · 2026-11-22
- OpenAI's first India hire Pragya Misra to lead business for India & South Asia — pragyamisra · 2026-11-22