OpenAIβs new reasoning technique alarms AI safety experts
2 September 2026 at 20:19
OpenAIβs new Astra model will use βrecurrent depth,β a technique that allows the model to operate outside of the sequential thinking that characterizes most reasoning models.