Watson apparently refines it's notion of the kinds of answers that are expected under a given category as it accumulates previous answers. The human contestants could exploit this by starting with the higher dollar questions. I'll be curious to see if they do.
There's a detailed chart of the performance over time of the system here.
Better late than never, a new open thread. Even with the discussion section, there are ideas or questions too short or inchoate to be worth a post.
This thread is for the discussion of Less Wrong topics that have not appeared in recent posts. If a discussion gets unwieldy, celebrate by turning it into a top-level post.