Previously, the Assistant was updated to not automatically scroll, so the user could read in-progress messages.
The new assistant (assistant.kagi.com) regresses this behavior: when the model finishes generating text, the user is forcibly scrolled to the end, even if they're in the middle of reading.
I seem to consistently get this if I'm scrolled part-way down a response when it finishes.
The assistant does not scroll on its own.
