Skip to content
← Back to explorer

Truth as a Compression Artifact in Language Model Training

Konstantin Krestnikov · Mar 12, 2026 · Citations: 0

Automatic Metrics Coding Math Pairwise Preference
Open arXiv Find Implementation RSS feed Shortlist (0)

How to use this page

Moderate trust

Use this for comparison and orientation, not as your only source.

Best use

Secondary protocol comparison source

What to verify

Validate the evaluation procedure and quality controls in the full paper before operational use.

Evidence quality

Moderate

Derived from extracted protocol signals and abstract evidence.

Abstract

Why do language models trained on contradictory data prefer correct answers? In controlled experiments with small transformers (3.5M--86M parameters), we show that this preference tracks the compressibility structure of errors rather than truth per se. We train GPT-2 style models on corpora where each mathematical problem appears with both correct and incorrect solutions -- a denoising design that directly models conflicting information about the same fact. When errors are random, models extract the correct signal with accuracy scaling from 65% to 85% with model size. When errors follow a coherent alternative rule system, accuracy drops to chance (~45--51%): the model cannot distinguish the false system from truth. A multi-rule experiment reveals a sharp crossover: a single coherent alternative rule eliminates truth bias entirely, but adding a second competing rule restores most of it (47%->78%), with continued growth through N=10 (88%). The same pattern reproduces on real Wikipedia text (71% vs 46%). We propose the Compression--Consistency Principle as an explanatory hypothesis: in these settings, gradient descent favors the most compressible answer cluster, not truth per se. Truth bias emerges only when falsehood is structurally incoherent. Whether this principle extends to large-scale pretraining remains an open question.

Should You Rely On This Paper?

This paper has useful evaluation signal, but protocol completeness is partial; pair it with related papers before deciding implementation strategy.

Best use

Secondary protocol comparison source

Use if you need

A secondary eval reference to pair with stronger protocol papers.

Main weakness

No major weakness surfaced.

Trust level

Moderate

Usefulness score

65/100 • Medium

Useful as a secondary reference; validate protocol details against neighboring papers.

Human Feedback Signal

Detected

Evaluation Signal

Detected

Usefulness for eval research

Moderate-confidence candidate

Extraction confidence 70%

If you are doing eval pipeline work, start here:

Human Eval Hub LLM-as-Judge Hub Pairwise Preference Hub Tool-Use Eval Hub

What We Could Verify

These are the protocol signals we could actually recover from the available paper metadata. Use them to decide whether this paper is worth deeper reading.

Human Feedback Types

strong

Pairwise Preference

Directly usable for protocol triage.

"Why do language models trained on contradictory data prefer correct answers?"

Evaluation Modes

strong

Automatic Metrics

Includes extracted eval setup.

"Why do language models trained on contradictory data prefer correct answers?"

Quality Controls

missing

Not reported

No explicit QC controls found.

"Why do language models trained on contradictory data prefer correct answers?"

Benchmarks / Datasets

missing

Not extracted

No benchmark anchors detected.

"Why do language models trained on contradictory data prefer correct answers?"

Reported Metrics

strong

Accuracy

Useful for evaluation criteria comparison.

"When errors are random, models extract the correct signal with accuracy scaling from 65% to 85% with model size."

Human Feedback Details

  • Uses human feedback: Yes
  • Feedback types: Pairwise Preference
  • Rater population: Not reported
  • Expertise required: Math, Coding

Evaluation Details

  • Evaluation modes: Automatic Metrics
  • Agentic eval: None
  • Quality controls: Not reported
  • Evidence quality: Moderate
  • Use this page as: Secondary protocol comparison source

Protocol And Measurement Signals

Benchmarks / Datasets

No benchmark or dataset names were extracted from the available abstract.

Reported Metrics

accuracy

Research Brief

Metadata summary

Why do language models trained on contradictory data prefer correct answers?

Based on abstract + metadata only. Check the source paper before making high-confidence protocol decisions.

Key Takeaways

  • Why do language models trained on contradictory data prefer correct answers?
  • In controlled experiments with small transformers (3.5M--86M parameters), we show that this preference tracks the compressibility structure of errors rather than truth per se.
  • We train GPT-2 style models on corpora where each mathematical problem appears with both correct and incorrect solutions -- a denoising design that directly models conflicting information about the same fact.

Researcher Actions

  • Compare this paper against nearby papers in the same arXiv category before using it for protocol decisions.
  • Validate inferred eval signals (Automatic metrics) against the full paper.
  • Use related-paper links to find stronger protocol-specific references.

Caveats

  • Generated from abstract + metadata only; no PDF parsing.
  • Signals below are heuristic and may miss details reported outside the abstract.

Research Summary

Contribution Summary

  • We introduce the Compression--Consistency Principle: next-token prediction favors hypotheses that allow shorter and more internally consistent descriptions of the training data.
  • 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).

Why It Matters For Eval

  • 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).

Researcher Checklist

  • Pass: Human feedback protocol is explicit

    Detected: Pairwise Preference

  • Pass: Evaluation mode is explicit

    Detected: Automatic Metrics

  • Gap: Quality control reporting appears

    No calibration/adjudication/IAA control explicitly detected.

  • Gap: Benchmark or dataset anchors are present

    No benchmark/dataset anchor extracted from abstract.

  • Pass: Metric reporting is present

    Detected: accuracy

Related Papers

Papers are ranked by protocol overlap, extraction signal alignment, and semantic proximity.

Get Started

Join the #1 Platform for AI Training Talent

Where top AI builders and expert AI Trainers connect to build the future of AI.
Self-Service
Post a Job
Post your project and get a shortlist of qualified AI Trainers and Data Labelers. Hire and manage your team in the tools you already use.
Create Account & Post a Job
Managed Service
For Large Projects
Done-for-You
We recruit, onboard, and manage a dedicated team inside your tools. End-to-end operations for large or complex projects.
Learn About Managed Service
For Freelancers
Join as an AI Trainer
Find AI training and data labeling projects across platforms, all in one place. One profile, one application process, more opportunities.
Join Now