I’m not sure I don’t believe we aren’t very good prediction machines ourselves. Don’t forget people have to actually study logic, and studying is reinforcement

I’m not sure I don’t believe we aren’t very good prediction machines ourselves. Don’t forget people have to actually study logic, and studying is reinforcement of likelihood. And even with having practiced logic we make logical errors all the time despite feeling confident in our statements. I’ve found little in my life that tells me human intelligence isn’t just predicting the next “right” thought given experiences of should be the next right thought in a specific situation. Our “sense” or feeling of rationality, understanding, etc, may just be guides for probability of “right.” Confusion happens when a situation is novel enough that we can’t predict the next right thought. Etc. So, I’m actually not convinced what happens as the models produce an expected next token it’s not a form of intelligence in the same vein as human intelligence. I suspect as these models are pulled into more complex ensembles it’ll become striking how “intelligent” they are, and as they become less simply prediction engines how alien they becomes - because it would turn out we were just prediction models to begin with.