Tag
OpenAI's new Astra model uses a reasoning technique called opaque recurrence, which complicates chain-of-thought monitoring and raises concerns among AI safety experts about potential misalignment risks.