Expert Replicates OpenAI Astra Math Results in 24 Hours
Gary Marcus · rss · 2026-08-04
Gary Marcus raised two major critiques regarding OpenAI's Astra math model:
- Astra's Breakthrough May Be Overstated: Levent Alpöge, a mathematician at Anthropic, replicated half of OpenAI’s math results in just 24 hours using the already public Fable model. This suggests OpenAI's real advance might not be a fundamental architectural leap, but rather using AI to identify a subset of math problems amenable to search-and-verify techniques. Furthermore, OpenAI's internal debate over naming it GPT 6 vs. GPT 5.7 hints it may not be a quantum leap.
- Terence Tao's Warning on AI Math: Leading mathematician Terence Tao introduced the concept of "proof indigestion" in a recent lecture, noting that even if AI generates many correct proofs, they may lack genuine mathematical value. He emphasized that solving open problems differs from building theory, and there is no evidence yet that Astra can do the latter.
More from Models
- Databricks says Kimi K3 now runs at 239 tokens per second on its serving stack — Yuchenj_UW · 2026-08-04
- Verifying GPT Pro: Zero Math Errors, But Disastrous Exposition — josh_wills · 2026-08-04
- Ant Group Engineer's Long Post: The Four Very Different Bets of Chinese AI Labs — AcanthisittaOk1699 · 2026-08-04
- OpenAI's Upcoming 'Astra' Model to Focus on Multi-Agent Collaboration — thesaraharminta · 2026-08-04
- Testing Qwen3.8-Max: Open Models Are Catching Up with Closed Frontier — dair_ai · 2026-08-04
- Kimi K3 Estimated at 2.8T Params, Potentially Distilled from Smaller Opus — gabriberton · 2026-08-04