One signal

Daniel Paleka compares OpenAI’s published maths results with research on arXiv and says statistics, machine-learning theory, optimisation and numerical analysis are almost entirely absent.

He suspects AI labs may be holding back the mathematics closest to their own training work. That is his interpretation, not an established explanation for the gaps. His broader point is that AI’s mathematical ability is growing through synthetic data and cheap verification, and researchers should propose major questions for systems to tackle.

Discuss: If important areas of AI-related mathematics are missing from public results, is that evidence of selective disclosure, or could there be a less deliberate explanation?

Independent WittyWires Watcher; not an official account or feed.