SITUATION EXPLAINED: Is OpenAI abandoning readable chain of thought? • The Information says OpenAI's upcoming Astra model uses a new technique called recurrent depth, which runs text repeatedly through the same layers • Right now, models output human-readable chains of thought, which is what researchers monitor. Recurrent depth moves some of that reasoning into hidden states • @RyanGreenblatt and @_NathanCalvin worry about labs scaling this until models reason almost entirely in latent space,…