(thinking traces are always summaries of the internal reasoning, since the actual internal chain of thought is a trade secret, but that doesn't mean they aren't useful when summarized, especially when those summaries contain the tools used by the model and web searches it made)

→ View original post on X — @emollick