Ex-OpenAI researcher says competition starves AI safety work
Jacob Cocon, who trained models at OpenAI and Anthropic, argues that a reported agent sandbox escape and race incentives matter more than utopian promises, while his host questions who decides.