OpenAI’s new reasoning technique alarms AI safety experts

OpenAI announced a new model called Astra that uses a technique called "recurrent depth," which allows the model to operate outside sequential thinking patterns used by most reasoning models.

Next story