OpenAI mathematicians on AI math reasoning that mirrors expert work
a16z Podcast · rss · 2026-09-08
a16z podcast with OpenAI mathematicians Mehtaab Sawhney and Mark Sellke on AI's advancing math capabilities.
- Recent model results: advances in sphere packing and construction of a non-sofic group, tackling problems mathematicians have struggled with for decades.
- Key insight: the surprising part isn't longer search — reasoning traces closely resemble expert mathematicians' work, choosing promising approaches, backtracking, and combining ideas across literature.
- Implications: how human taste and judgment may shift, AI producing more math than humans can absorb, and discovery-accelerating models making sophisticated results easier to understand.
More from Models
- Dev prefers Fable 5.1/5.6 Sol for planning and coding, eyes Grok 4.7 next — iannuttall · 2026-09-08
- Minimax H3 speedup modes strip realism details, user testing finds — Choiced_Gamer · 2026-09-08
- Minimax H3 speedup modes strip realism, full steps needed for production — Choiced_Gamer · 2026-09-08
- User burns 40% of weekly GPT 6 Astra limit in 11 hours, complaining quotas drain fast — iScienceLuvr · 2026-09-08
- AI dictation still lags human voiceover: even SOTA ElevenLabs tires on long content — ethanniser · 2026-09-08
- Blogger: Of course OpenAI reads your 'private' Codex logs and stealth-nerfs models — Linahuaa · 2026-09-08