OpenAI’s new model, Astra, will employ a technique described as “recurrent depth,” which lets the model reason outside the sequential thinking that characterizes most current reasoning models. According to the report, this approach has raised concerns among AI safety experts.
Why it matters
Reasoning models typically follow a sequential process, and a method that operates differently could change how these systems arrive at conclusions. The reported unease from safety experts signals that the technique’s implications are not yet fully understood.
Who should care
AI safety researchers and those tracking developments in reasoning models will want to follow how recurrent depth is implemented and evaluated in Astra.