Okay, please educate me on this progress and how LLMs are different to generating the next most statistically probable token.

How about _you_ educate us how _you_ are any different than generating the most statistically probable token?