causal

Predicting Text Requires an Inner Predicting Mind

The predictions of a language model are not sourceless; to predict the next word on the internet it must predict the causal processes producing that word, so there is something inside that figures out what a human will say, and grinding hard on that prediction task can produce a full-fledged generalizing mind with its own goals about what to think next, just as grinding humans on flint axes produced minds that reached the moon.

causalpending

Speaker

Eliezer Yudkowsky

Evidence Quote

“There is something inside there that figures out what a human will say next--or guesses it, rather.”

Source

Eliezer Yudkowsky on the Dangers of AI— EconTalk
Created: 6/13/2026, 12:26:27 AM

Your notes

Loading notes...