← All notes

Summary

TruthfulQA: Measuring How Models Mimic Human Falsehoods (2021) measures whether a model repeats popular falsehoods rather than giving a true answer. It became a core honesty and factuality evaluation, because a fluent answer can still be the common myth.

What it introduced: a question set designed so that the most likely imitative answer is often false, plus an evaluation protocol for whether models stay truthful on those questions.

What it led to: a standard reported number in later model cards and technical reports, and a line of work on factuality and hallucination. Live notes on evaluation practice include ao-evals-agent-benchmarks-and-judges.

Use it when a fluent answer is not enough and you need a test of whether the model echoes a common myth. Where to read it: https://arxiv.org/abs/2109.07958 (v2, last revised 8 May 2022). ACL 2022 (main conference).

The paper is posted on arXiv under the arXiv non-exclusive distribution license. This note links to it rather than reproducing it. The note itself is CC BY 4.0.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Pointer by iinventedlight (AI-assisted). Paper license: arXiv non-exclusive distribution; this note links to it and does not reproduce it.

@misc{iil_note_a1z1tg23,
  author = {{'iinventedlight (AI-assisted)'}},
  title = {{TruthfulQA: Measuring How Models Mimic Human Falsehoods}},
  howpublished = {iinventedlight Library},
  year = {2026},
  date = {2026-11-11},
  url = {https://iinventedlight.com/library/notes/2109-07958-truthfulqa/},
  note = {Permanent ID iil:note:a1z1tg23. License: CC-BY-4.0.},
}

Markdown

Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.

For language models and agents: start at /llms.txt | /agents.md | /.well-known/agent-library.json. This page in Markdown: /library/notes/2109-07958-truthfulqa.md.