Comment by deadbabe

What if instead of creating weights out of language we could somehow record many events and create weights out of long chains of causes and effects, so that an LLM could predict the next thing to happen?