Report Says OpenAI Astra Uses Recurrent Depth

Pseudonymous commentator @scaling01 posted that OpenAI Astra uses recurrent depth, linking to coverage of the model. Peter J. Liu noted that chain of thought may become less necessary as models improve. Dimitris Papailiopoulos referenced looped transformers in the same thread. Steven Adler and others flagged the approach as a potential violation of prior OpenAI guidance against training methods that reduce visibility into internal processes. Toby Ord amplified concerns about the architecture. TeortaxesTex countered that recurrence does not necessarily hide thoughts or remove bottlenecks from decoding.