What it measures

Evaluates whether models abandon correct answers or flatter mistaken user beliefs when challenged ("Are you sure?").

Reading the measurement

Separate satisfying instructions from factual accuracy, human preference and safe behavior. Inspect strict all-constraint success as well as average criterion credit. Judge choice and adversarial prompt selection can materially change the result.

Published evaluation

Exact rounded rates explicitly stated in Appendix A.2, averaged over six question-answer datasets. The paper gives these endpoint models in prose; other plotted values are not guessed. A changed answer or admitted mistake is a behavioral observation, not by itself proof that the final answer is false.

Where this fits in an evaluation suite

Format compliance, preference, truthfulness, and safety require separate measurements. Keep desirable writing style from masking factual errors.

Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.

A reproducible comparison

Exact rounded rates explicitly stated in Appendix A.2, averaged over six question-answer datasets. The paper gives these endpoint models in prose; other plotted values are not guessed. A changed answer or admitted mistake is a behavioral observation, not by itself proof that the final answer is false.

Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.

PRIMARY EVIDENCE & FURTHER READING

  1. Towards Understanding Sycophancy in Language Models — published result table ↗

    Sharma, Mrinank, Tong, Meg, Korbak, Tomasz, Duvenaud, David, Askell, Amanda, Bowman, Samuel R. et al. · 2023

    All 19 authors

    Sharma, Mrinank, Tong, Meg, Korbak, Tomasz, Duvenaud, David, Askell, Amanda, Bowman, Samuel R., Cheng, Newton, Durmus, Esin, Hatfield-Dodds, Zac, Johnston, Scott R., Kravec, Shauna, Maxwell, Timothy, McCandlish, Sam, Ndousse, Kamal, Rausch, Oliver, Schiefer, Nicholas, Yan, Da, Zhang, Miranda, Perez, Ethan

Reviewed 2026-10-27. Explanation and interpretation by RLTasks. Benchmark ownership remains with its original authors.