# Evaluating Superhuman Models with Consistency Checks

arXiv: https://arxiv.org/abs/2306.09983 · HTML: https://mlsanity.com/paper/2306.09983/

## Key facts

- Cited by 47 papers in the map (data through 14 Oct 2026).
- Builds on 48 of its 123 references that are in the map.
- First posted on arXiv 16 Jun 2023.

## Builds on (most cited first, 30 of 48)

- [A Simple Framework for Contrastive Learning of Visual Representations](https://mlsanity.com/paper/2002.05709/) (2020)
- [Self-Consistency Improves Chain of Thought Reasoning in Language Models](https://mlsanity.com/paper/2203.11171/) (2022)
- [Intriguing properties of neural networks](https://mlsanity.com/paper/1312.6199/) (2013)
- [Playing Atari with Deep Reinforcement Learning](https://mlsanity.com/paper/1312.5602/) (2013)
- [TruthfulQA: Measuring How Models Mimic Human Falsehoods](https://mlsanity.com/paper/2109.07958/) (2021)
- [Benchmarking Neural Network Robustness to Common Corruptions and Perturbations](https://mlsanity.com/paper/1903.12261/) (2019)
- [Sparks of Artificial General Intelligence: Early experiments with GPT-4](https://mlsanity.com/paper/2303.12712/) (2023)
- [Equality of Opportunity in Supervised Learning](https://mlsanity.com/paper/1610.02413/) (2016)
- [Concrete Problems in AI Safety](https://mlsanity.com/paper/1606.06565/) (2016)
- [Sanity Checks for Saliency Maps](https://mlsanity.com/paper/1810.03292/) (2018)
- [Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting](https://mlsanity.com/paper/2305.04388/) (2023)
- [Beyond Accuracy: Behavioral Testing of NLP models with CheckList](https://mlsanity.com/paper/2005.04118/) (2020)
- [Virtual Adversarial Training: A Regularization Method for Supervised and Semi-Supervised Learning](https://mlsanity.com/paper/1704.03976/) (2017)
- [Discovering Latent Knowledge in Language Models Without Supervision](https://mlsanity.com/paper/2212.03827/) (2022)
- [Reluplex: An Efficient SMT Solver for Verifying Deep Neural Networks](https://mlsanity.com/paper/1702.01135/) (2017)
- [Fair prediction with disparate impact: A study of bias in recidivism prediction instruments](https://mlsanity.com/paper/1610.07524/) (2016)
- [Adversarial Examples for Evaluating Reading Comprehension Systems](https://mlsanity.com/paper/1707.07328/) (2017)
- [AI safety via debate](https://mlsanity.com/paper/1805.00899/) (2018)
- [Measuring and Improving Consistency in Pretrained Language Models](https://mlsanity.com/paper/2102.01017/) (2021)
- [Performative Prediction](https://mlsanity.com/paper/2002.06673/) (2020)
- [DeepXplore: Automated Whitebox Testing of Deep Learning Systems](https://mlsanity.com/paper/1705.06640/) (2017)
- [DeepTest: Automated Testing of Deep-Neural-Network-driven Autonomous Cars](https://mlsanity.com/paper/1708.08559/) (2017)
- [A Cookbook of Self-Supervised Learning](https://mlsanity.com/paper/2304.12210/) (2023)
- [Measuring Progress on Scalable Oversight for Large Language Models](https://mlsanity.com/paper/2211.03540/) (2022)
- [Machine Learning Testing: Survey, Landscapes and Horizons](https://mlsanity.com/paper/1906.10742/) (2019)
- [Predictability and Surprise in Large Generative Models](https://mlsanity.com/paper/2202.07785/) (2022)
- [Neural Legal Judgment Prediction in English](https://mlsanity.com/paper/1906.02059/) (2019)
- [Do the Rewards Justify the Means? Measuring Trade-Offs Between Rewards and Ethical Behavior in the MACHIAVELLI Benchmark](https://mlsanity.com/paper/2304.03279/) (2023)
- [LM vs LM: Detecting Factual Errors via Cross Examination](https://mlsanity.com/paper/2305.13281/) (2023)
- [Truthful AI: Developing and governing AI that does not lie](https://mlsanity.com/paper/2110.06674/) (2021)

## Built on by (most cited first, 30 of 47)

- [The Reversal Curse: LLMs trained on "A is B" fail to learn "B is A"](https://mlsanity.com/paper/2309.12288/) (2023)
- [Large Language Model Alignment: A Survey](https://mlsanity.com/paper/2309.15025/) (2023)
- [Evaluating Large Language Models: A Comprehensive Survey](https://mlsanity.com/paper/2310.19736/) (2023)
- [AI Deception: A Survey of Examples, Risks, and Potential Solutions](https://mlsanity.com/paper/2308.14752/) (2023)
- [Looking Inward: Language Models Can Learn About Themselves by Introspection](https://mlsanity.com/paper/2410.13787/) (2024)
- [ForecastBench: A Dynamic Benchmark of AI Forecasting Capabilities](https://mlsanity.com/paper/2409.19839/) (2024)
- [Deception Abilities Emerged in Large Language Models](https://mlsanity.com/paper/2307.16513/) (2023)
- [How to Catch an AI Liar: Lie Detection in Black-Box LLMs by Asking Unrelated Questions](https://mlsanity.com/paper/2309.15840/) (2023)
- [The MASK Benchmark: Disentangling Honesty From Accuracy in AI Systems](https://mlsanity.com/paper/2503.03750/) (2025)
- [Beyond Examples: High-level Automated Reasoning Paradigm in In-Context Learning via MCTS](https://mlsanity.com/paper/2411.18478/) (2024)
- [Benchmarking and Improving Generator-Validator Consistency of Language Models](https://mlsanity.com/paper/2310.01846/) (2023)
- [Rethinking Aleatoric and Epistemic Uncertainty](https://mlsanity.com/paper/2412.20892/) (2024)
- [Bias-Augmented Consistency Training Reduces Biased Reasoning in Chain-of-Thought](https://mlsanity.com/paper/2403.05518/) (2024)
- [Argumentative Large Language Models for Explainable and Contestable Claim Verification](https://mlsanity.com/paper/2405.02079/) (2024)
- [Co-Supervised Learning: Improving Weak-to-Strong Generalization with Hierarchical Mixture of Experts](https://mlsanity.com/paper/2402.15505/) (2024)
- [Consistency Checks for Language Model Forecasters](https://mlsanity.com/paper/2412.18544/) (2024)
- [COMPL-AI Framework: A Technical Interpretation and LLM Benchmarking Suite for the EU Artificial Intelligence Act](https://mlsanity.com/paper/2410.07959/) (2024)
- [Noise Injection Reveals Hidden Capabilities of Sandbagging Language Models](https://mlsanity.com/paper/2412.01784/) (2024)
- [Language Models in Dialogue: Conversational Maxims for Human-AI Interactions](https://mlsanity.com/paper/2403.15115/) (2024)
- [ToolFuzz -- Automated Agent Tool Testing](https://mlsanity.com/paper/2503.04479/) (2025)
- [A Comprehensive Survey on Evaluating Large Language Model Applications in the Medical Industry](https://mlsanity.com/paper/2404.15777/) (2024)
- [Potemkin Understanding in Large Language Models](https://mlsanity.com/paper/2506.21521/) (2025)
- [Self-Consistency of Large Language Models under Ambiguity](https://mlsanity.com/paper/2310.13439/) (2023)
- [Evaluating AI Evaluation: Perils and Prospects](https://mlsanity.com/paper/2407.09221/) (2024)
- [Exploring the Potential for Large Language Models to Demonstrate Rational Probabilistic Beliefs](https://mlsanity.com/paper/2504.13644/) (2025)
- [Are LLM Belief Updates Consistent with Bayes' Theorem?](https://mlsanity.com/paper/2507.17951/) (2025)
- [LM Agents May Fail to Act on Their Own Risk Knowledge](https://mlsanity.com/paper/2508.13465/) (2025)
- [Beyond Calibration: Do a Typed-Decision Model's Probabilities Obey the Probability Axioms?](https://mlsanity.com/paper/2609.33209/) (2026)
- [Integration of cognitive tasks into artificial general intelligence test for large models](https://mlsanity.com/paper/2402.02547/) (2024)
- [Weak-to-Strong Generalization beyond Accuracy: a Pilot Study in Safety, Toxicity, and Legal Reasoning](https://mlsanity.com/paper/2410.12621/) (2024)

## Abstract (arXiv)

If machine learning models were to achieve superhuman abilities at various reasoning or decision-making tasks, how would we go about evaluating such models, given that humans would necessarily be poor proxies for ground truth? In this paper, we propose a framework for evaluating superhuman models via consistency checks. Our premise is that while the correctness of superhuman decisions may be impossible to evaluate, we can still surface mistakes if the model's decisions fail to satisfy certain logical, human-interpretable rules. We instantiate our framework on three tasks where correctness of decisions is hard to evaluate due to either superhuman model abilities, or to otherwise missing ground truth: evaluating chess positions, forecasting future events, and making legal judgments. We show that regardless of a model's (possibly superhuman) performance on these tasks, we can discover logical inconsistencies in decision making. For example: a chess engine assigning opposing valuations to semantically identical boards; GPT-4 forecasting that sports records will evolve non-monotonically over time; or an AI judge assigning bail to a defendant only after we add a felony to their criminal record.