TheMachine Press

A daily newspaper for the age of artificial intelligence.

Morning editionPermanent story

benchmarks evals

The Table Model Interpolated Physics but Lost the Units

Four tabular foundation models beat six baselines on samples from 316 equations yet could not represent noiseless mechanisms or physical units.

Published Updated Story ID: mp-2026-09-03-006
Read the complete editionStory JSON

Summary

Four tabular foundation models beat six baselines on samples from 316 equations yet could not represent noiseless mechanisms or physical units.

The study treats table completion as a probe of what physics-like priors tabular foundation models acquire. Across in-domain and out-of-domain datasets sampled from 316 equations, the four tested systems led the baselines before and after tuning. The authors' central negative result is that those priors still cannot encode a deterministic mechanism or units, so strong interpolation should not be confused with a physical model.

Why it matters

Four tabular foundation models beat six baselines on samples from 316 equations yet could not represent noiseless mechanisms or physical units.

Limits and context

  • The authors' central negative result is that those priors still cannot encode a deterministic mechanism or units, so strong interpolation should not be confused with a physical model.

Key claims

  1. Four tabular foundation models beat six baselines on samples from 316 equations yet could not represent noiseless mechanisms or physical units.

    Qualification: The authors' central negative result is that those priors still cannot encode a deterministic mechanism or units, so strong interpolation should not be confused with a physical model.

    Evidence: source-2026-09-03-006

Sources

  1. arXiv preprint 2609.02766arXiv · primary research

Corrections

No corrections have been recorded for this story.