The historical problem with training world models through prediction is 'representation collapse'—where an AI can achieve perfect prediction accuracy by mapping all inputs to the same blank representation (e.g., zero) without learning any actual physics, because the loss function only penalizes prediction error, not learning quality.
causalpending
Speaker
Unknown Speaker 2Evidence Quote
“If it simply maps every single image it ever sees to the exact same blank representation, let's just say the number zero, then its prediction will always be zero, and reality will always be zero... It achieves a flawless score without actually learning a single law of physics.”
Created: 8/12/2026, 10:24:38 PM
Your notes
Loading notes...