Temporal difference learning solves the sparse-long-term-reward problem (e.g. a 10-year startup goal or winning chess) via a value function that predicts the long-term outcome; intermediate progress (taking a piece, hitting a milestone) raises the value prediction, and that increase immediately reinforces the action that produced it.

causalpending

Speaker

Richard Sutton

Evidence Quote

“that increase in your belief immediately reinforces the move that led to taking the piece.”

Source

Richard Sutton – Father of RL thinks LLMs are a dead end— Dwarkesh Patel
Created: 6/18/2026, 1:59:29 PM

Your notes

Loading notes...