Mathematician alleges OpenAI's Astra may have trained on his unpublished soficity work
basedjensen · x · 2026-09-10
Mathematician Valerio Capraro relays Andreas Thom's detailed Mastodon post presenting evidence that OpenAI may have trained Astra on conversations in which Thom and Gábor Kun worked on Gromov's soficity conjecture — one of ten problems OpenAI later announced Astra solved.
- Thom is a leading expert on sofic/hyperlinear groups who spent two decades on the problem
- If accurate, this goes beyond attribution disputes into whether unpublished research conversations were used as training data
- Unverified, but adds a major allegation to the controversy around Astra's math breakthroughs
Related event: Mathematicians Accuse OpenAI of Training Astra on Unpublished Work(5 posts)→
More from Models
- $20-tier user reports Astra Light exhausts usage in 5-6 prompts, unlike Sol — HousingLatter6637 · 2026-09-10
- V4.1 Flash KV cache is 437x smaller than V1, easing China's memory bottleneck — ChrisGPT · 2026-09-10
- DeepSeek V4.1 Flash rumored: 552B MoE with asymmetric CED architecture, beats flagship — _AndrewZhao · 2026-09-10
- DeepSeek-V4.1-Flash Hits the API: 1/8 the SSD Cache, V4-Pro Being Phased Out — deepseek_ai · 2026-09-10
- DeepSeek Launches V4.1-Flash, Smallest Model in New Architecture Family with Native Vision — deepseek_ai · 2026-09-10
- Google releases TimesFM-3, a 330M multivariate zero-shot forecasting foundation model — AxSaucedo · 2026-09-10