OpenAI's new reasoning technique alarms AI safety experts

A new reasoning technique reportedly to be used in OpenAI's new Astra model, which allows it to step outside of sequential thought structures, has drawn criticism from AI safety experts.

Recurrent Depth Technique

OpenAI's new Astra model is reportedly set to use a technique called recurrent depth, which allows it to operate outside the sequential thinking that characterizes traditional reasoning models.

Experts' Concerns

This approach, also referred to as non-transparent iteration, has caused serious concern among AI safety experts because it can make it harder to monitor the model's chain of thought.

Chain of Thought Traceability

Normally, a model's chain of thought shows the sequential steps taken during problem-solving, but with this new method, the same query is processed repeatedly in a loop, leaving a less readable trail.

Future Regulations

Experts point out that regulations may be necessary to prevent such techniques from triggering a dangerous race among laboratories.