Mathematician Claims AI Labs Withhold Internal Results

Thomas Bloom, a Royal Society University Research Fellow at the University of Manchester, posted that it is widely known both OpenAI and Anthropic possess many results proved by internal models beyond those already announced. He asks whether keeping the findings unreleased is a good or bad thing. The post ends with 1/ and forms the sole source line supplied. The statement attributes the existence of the unreleased results to common knowledge rather than direct confirmation from the companies.