DoubleChecks

What was claimed

Reasoning models produce fractals when solving hard problems (Sudoku, math, ARC-AGI); nonlinear dynamics can now probe the 'thinking processes' of these models

Our verdict

Needs caution

The paper shows reasoning traces can be treated as dynamical systems and analyzed with nonlinear-dynamics concepts. But saying this probes the models' actual 'thinking processes' goes beyond what the source explicitly proves. The provided paper mentions Sudoku, maze solving, visual puzzles, and mathematical logic, but ARC-AGI is not clearly named in the quoted source. The claim may be true in related work, but it is not directly confirmed here. (Only 2 of 3 AI systems responded.)

Combined assessment44 sourcesChecked Sep 9, 2026

Why this verdict

  • Nonlinear dynamics can now probe the 'thinking processes' of these models.

    Misleading
    One assessment reached this conclusion

  • Reasoning models produce fractals when solving ARC-AGI.

    Can’t verify
    One assessment reached this conclusion

  • Reasoning models produce fractals when solving hard problems such as Sudoku and math.

    Verified
    One assessment reached this conclusion

Evidence reviewed

44 sources

Perplexity and Claude sources are combined here.

arxiv.orgarxiv.orgarxiv.orgopenai.comdeeplearning.aiarxiv.orgarxiv.orgmindstudio.ailinkedin.comarcprize.org

Individual assessments

Assessment summary

  • The response is broadly consistent with current research, but it is overstated.
  • A recent paper does report fractal basins in reasoning models on tasks such as Sudoku and math-like logic, and frames reasoning traces as a dynamical-system object that can be analyzed with nonlinear dynamics arxiv.org.
  • However, the claim that this applies to ARC-AGI is not clearly supported by the provided source, and the phrasing “nonlinear dynamics can now probe the 'thinking processes' of these models” is more interpretive than directly established arxiv.org.

Claims reviewed

Nonlinear dynamics can now probe the 'thinking processes' of these models.

Misleading
The paper shows reasoning traces can be treated as dynamical systems and analyzed with nonlinear-dynamics concepts arxiv.org. But saying this probes the models' actual 'thinking processes' goes beyond what the source explicitly proves.

This claim may mislead readers without added context.

Reasoning models produce fractals when solving ARC-AGI.

Can’t verify
The provided paper mentions Sudoku, maze solving, visual puzzles, and mathematical logic, but ARC-AGI is not clearly named in the quoted source arxiv.org. The claim may be true in related work, but it is not directly confirmed here.

Reasoning models produce fractals when solving hard problems such as Sudoku and math.

Verified
The cited paper explicitly says leading reasoning models show fractal basins on harder tasks, including Sudoku and mathematical logic arxiv.org.

Assessment summary

  • Reasoning models are dynamical systems with fractal basins, with fractality increasing with task difficulty across diverse tasks like Sudoku and maze solving, visual puzzles, and mathematical logic.
  • Reasoning models exhibit transient chaos, a physical consequence of the computational complexity of difficult tasks.
  • The claim accurately reflects recent peer-reviewed research findings.
ChatGPTDid not respond

Check your own claim

Paste any statement, headline, or AI answer. Independent assessments check it and compare sources.

Share this result