Search papers, labs, and topics across Lattice.
The paper investigates why language models sometimes favor correct statements even when trained on noisy data, proposing the Compression-Consistency Principle: models prefer hypotheses enabling shorter, more consistent data descriptions. Using GPT-2-style transformers on synthetic math corpora with controlled errors, the study demonstrates that models strongly prefer correct completions when errors are random but lose this preference when errors form a coherent, incorrect system. Embedding verification steps restores preference for correctness, suggesting that "truth bias" is a byproduct of compression and consistency, not an intrinsic drive.
Language models' apparent "truth bias" is often just a side effect of favoring easily compressible, internally consistent patterns, not an inherent grasp of truth.
Why do language models sometimes prefer correct statements even when trained on mixed-quality data? We introduce the Compression--Consistency Principle: next-token prediction favors hypotheses that allow shorter and more internally consistent descriptions of the training data. Truth bias emerges only when false alternatives are structurally harder to compress. We test this using small GPT-2-style character-level transformers (3.5M--86M parameters) on synthetic math corpora with controlled mixtures of correct and incorrect rules. In the random-error setting, models strongly prefer correct completions in paired evaluation: 83.1% accuracy at balanced data and 67.0% even when correct rules appear in only 10% of the corpus. Replacing random errors with a coherent but mathematically incorrect rule system largely eliminates the preference (near-chance accuracy). In a more natural-language-like synthetic world, the effect is weaker but still present (57.7%). Additional experiments show that embedding verification steps can restore preference for correctness even at small scale, while increasing the number of consistent rules produces a graded improvement in accuracy. Our results suggest that what appears as a"truth bias"is largely a side effect of compression pressure and preference for internal consistency, rather than an intrinsic drive toward truth. Full code and data are available at https://github.com/Rai220/compression-drives-truth.
