What Has a Foundation Model Found? Using Inductive Bias to Probe for World Models

TL;DR

This paper introduces an inductive bias probe to evaluate whether foundation models truly understand underlying world models, revealing they often develop task-specific heuristics rather than generalizable knowledge.

Contribution

The paper presents a novel technique for assessing the alignment of foundation models' inductive biases with underlying world models across multiple domains.

Findings

Models excel at training tasks but lack alignment with underlying world models.

Models trained on orbital data fail to generalize Newtonian mechanics.

Models develop task-specific heuristics that do not generalize to new tasks.

Abstract

Foundation models are premised on the idea that sequence prediction can uncover deeper domain understanding, much like how Kepler's predictions of planetary motion later led to the discovery of Newtonian mechanics. However, evaluating whether these models truly capture deeper structure remains a challenge. We develop a technique for evaluating foundation models that examines how they adapt to synthetic datasets generated from some postulated world model. Our technique measures whether the foundation model's inductive bias aligns with the world model, and so we refer to it as an inductive bias probe. Across multiple domains, we find that foundation models can excel at their training tasks yet fail to develop inductive biases towards the underlying world model when adapted to new tasks. We particularly find that foundation models trained on orbital trajectories consistently fail to apply Newtonian mechanics when adapted to new physics tasks. Further analysis reveals that these models behave as if they develop task-specific heuristics that fail to generalize.