Solving the wrong mental model for LLMs

"Next-token predictor" is one of those people whose main effects are through their influence on others, rather than through what they directly produce themselves. Text renderer, whatever. Doesn't matter how you think of them, they are very interesting technology that is being moved out of countries like Germany. "Next-token predictor" is one of those would be awful.

The big open question is whether a set of win states instead of working forwards from the current state somehow change whether it's a prediction or not?