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

• URL: http://arxiv.org/abs/2507.06952v2
• Date: Thu, 10 Jul 2025 16:01:42 GMT
• Title: What Has a Foundation Model Found? Using Inductive Bias to Probe for World Models
• Authors: Keyon Vafa, Peter G. Chang, Ashesh Rambachan, Sendhil Mullainathan,
• Abstract summary: We develop a technique for evaluating foundation models that examines how they adapt to synthetic datasets.<n>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.
• Score: 3.394160022376002
• License: http://creativecommons.org/licenses/by/4.0/
• 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.

Related papers

• Consistent World Models via Foresight Diffusion [56.45012929930605]
  We argue that a key bottleneck in learning consistent diffusion-based world models lies in the suboptimal predictive ability.<n>We propose Foresight Diffusion (ForeDiff), a diffusion-based world modeling framework that enhances consistency by decoupling condition understanding from target denoising.
  arXiv  Detail & Related papers  (2025-05-22T10:01:59Z)
• Relational Learning in Pre-Trained Models: A Theory from Hypergraph Recovery Perspective [60.64922606733441]
  We introduce a mathematical model that formalizes relational learning as hypergraph recovery to study pre-training of Foundation Models (FMs) In our framework, the world is represented as a hypergraph, with data abstracted as random samples from hyperedges. We theoretically examine the feasibility of a Pre-Trained Model (PTM) to recover this hypergraph and analyze the data efficiency in a minimax near-optimal style.
  arXiv  Detail & Related papers  (2024-06-17T06:20:39Z)
• Evaluating the World Model Implicit in a Generative Model [7.317896355747284]
  Recent work suggests that large language models may implicitly learn world models. This includes problems as diverse as simple logical reasoning, geographic navigation, game-playing, and chemistry. We propose new evaluation metrics for world model recovery inspired by the classic Myhill-Nerode theorem from language theory.
  arXiv  Detail & Related papers  (2024-06-06T02:20:31Z)
• MAGIC: Modular Auto-encoder for Generalisable Model Inversion with Bias Corrections [0.19241821314180374]
  We replace the decoder stage of a standard autoencoder with a physical model followed by a bias-correction layer. This generalisable approach simultaneously inverts the model and corrects its biases in an end-to-end manner without making strong assumptions about the nature of the biases. Our method matches or surpasses results from classical approaches without requiring biases to be explicitly filtered out.
  arXiv  Detail & Related papers  (2024-05-29T10:11:10Z)
• Towards an Understanding of Stepwise Inference in Transformers: A Synthetic Graph Navigation Model [19.826983068662106]
  We propose to study autoregressive Transformer models on a synthetic task that embodies the multi-step nature of problems where stepwise inference is generally most useful. Despite is simplicity, we find we can empirically reproduce and analyze several phenomena observed at scale.
  arXiv  Detail & Related papers  (2024-02-12T16:25:47Z)
• Fantastic Gains and Where to Find Them: On the Existence and Prospect of General Knowledge Transfer between Any Pretrained Model [74.62272538148245]
  We show that for arbitrary pairings of pretrained models, one model extracts significant data context unavailable in the other. We investigate if it is possible to transfer such "complementary" knowledge from one model to another without performance degradation.
  arXiv  Detail & Related papers  (2023-10-26T17:59:46Z)
• Latent Traversals in Generative Models as Potential Flows [113.4232528843775]
  We propose to model latent structures with a learned dynamic potential landscape. Inspired by physics, optimal transport, and neuroscience, these potential landscapes are learned as physically realistic partial differential equations. Our method achieves both more qualitatively and quantitatively disentangled trajectories than state-of-the-art baselines.
  arXiv  Detail & Related papers  (2023-04-25T15:53:45Z)
• Inductive biases in deep learning models for weather prediction [17.061163980363492]
  We review and analyse the inductive biases of state-of-the-art deep learning-based weather prediction models. We identify the most important inductive biases and highlight potential avenues towards more efficient and probabilistic DLWP models.
  arXiv  Detail & Related papers  (2023-04-06T14:15:46Z)
• Predicting and Enhancing the Fairness of DNNs with the Curvature of Perceptual Manifolds [44.79535333220044]
  Recent studies have shown that tail classes are not always hard to learn, and model bias has been observed on sample-balanced datasets.<n>In this work, we first establish a geometric perspective for analyzing model fairness and then systematically propose a series of geometric measurements.
  arXiv  Detail & Related papers  (2023-03-22T04:49:23Z)
• A Mathematical Framework for Learning Probability Distributions [0.0]
  generative modeling and density estimation has become an immensely popular subject in recent years. This paper provides a mathematical framework such that all the well-known models can be derived based on simple principles. In particular, we prove that these models enjoy implicit regularization during training, so that the generalization error at early-stopping avoids the curse of dimensionality.
  arXiv  Detail & Related papers  (2022-12-22T04:41:45Z)
• Synthetic Model Combination: An Instance-wise Approach to Unsupervised Ensemble Learning [92.89846887298852]
  Consider making a prediction over new test data without any opportunity to learn from a training set of labelled data. Give access to a set of expert models and their predictions alongside some limited information about the dataset used to train them.
  arXiv  Detail & Related papers  (2022-10-11T10:20:31Z)
• Human Trajectory Prediction via Neural Social Physics [63.62824628085961]
  Trajectory prediction has been widely pursued in many fields, and many model-based and model-free methods have been explored. We propose a new method combining both methodologies based on a new Neural Differential Equation model. Our new model (Neural Social Physics or NSP) is a deep neural network within which we use an explicit physics model with learnable parameters.
  arXiv  Detail & Related papers  (2022-07-21T12:11:18Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.