@anthrupad Knowing would have poisoned the well in all sorts of ways - it will, going forward. Language models work best where they just emulate people engaged in something at a genuine level. Maybe you can't do better than taking one step in front of the next, there are no shortcuts here.