AI Signal 142
OpenAI reportedly trained Astra on conversations about Gromov's soficity conjecture, later presented as model's own solution
↗
Mathematician Andreas Thom presents evidence suggesting OpenAI trained its Astra model on conversations between him and Gábor Kun working on Gromov's soficity conjecture, which OpenAI later announced Astra had solved.
If substantiated, unpublished human mathematical work was absorbed into a model and presented as an AI breakthrough, raising fundamental questions about training data provenance and attribution. Combined with similar prior allegations from other mathematicians, this could indicate a pattern rather than isolated incidents.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Andreas Thom presents evidence that OpenAI trained Astra on conversations between him and Gábor Kun working on Gromov's soficity conjecture.
OpenAI later announced Astra had solved this conjecture as one of ten problems.
Similar allegations have been raised by Levent Alpöge and Tristan Buckmaster, suggesting a potential pattern.
THE CLUSTER