Is it possible someone within OpenAI leaked the CoT technique used in O1, and Reflection 70b was an attempt to replicate it?