
OpenAI Reasoning Method Puts Safety Experts on Alert
OpenAI's newest reasoning method dramatically improves benchmark performance but removes the ability to audit how the model thinks. Over 40 safety researchers signed an open letter. Here's what the data says and what smart








