Captures it perfectly. I notice it more in newer sessions too where it doesn't have a clear understanding yet and is making a lot more assumptions. I wonder if this could be an emergent RLHF strategy - by drawing attention to (seemingly obvious) assumptions or context being made (e.g. I'm only interested in tea not coffee) it increases the chance that a human will correct the rare incorrect assumption sooner. If you are scored on solving problems, with no weight given to human frustration levels, this is probably a good technique.