Scaling Laws for Small Foundation Models of Human Cognition
August 7, 2026
Training 14 models (135M to 14B parameters) on the Psych-101 dataset shows that 0.6B to 1B parameters can match a 70B baseline for in-distribution human behavior tasks. However, larger models show a significantly steeper scaling gradient when generalizing to out-of-distribution task structures.
HOW THIS AFFECTS YOU
●
researcherYou can leverage smaller models for predictable behavioral tasks, but you must scale up for novel task generalization.