As a memory neuroscientist, I enjoyed the shoutout here to episodic memory. It strikes me, however, that a feature that I've noticed when observing "reasoning" models is that they may explicitly search for evidence for intermediate pieces of their responses. If we're following the "remember"/"know" distinction developed by Squire and others, perhaps the more apt analogy might be that a singly pass through an LLM is similar to a "I know this" result, prone to hallucination, conflation, etc., and the multipass reasoning model might be more akin to the "I remember this" result, where primary evidence serves as a substrate for the response?

The terms are arbitrary and don't relate to human memory. Using the term "I know this" is a post-hoc retrofit to a process exclusively accessed wordlessly. Just as "I remember this" does not access that process, but rather comments on the aftermath pretty unscientifically, like a sportscaster trying to read a pitcher's mind.

“We refute (based on empirical evidence) claims that humans use linguistic representations to think.” Ev Fedorenko Language Lab MIT 2024