[rsynnott]

Tay was a bit different, in that it was actually training off its interactions with users (ie it had a constant ongoing training process). LLMs don't do that.