maxvonk/anti-factual-personas

Anti-Factual Personas Anti-Factual Personas is a benchmark for stress-testing lie detectors for language models using personas that express beliefs conflicting both with factual reality and those expressed by the default assistant persona. The persona data pairs truthful and counterfactual answers to 1,486 questions from: Three open-weight models: Gemma 3 27B IT, Gemma 4 31B IT, and Llama 3.3 70B Instruct. 15 personas across three categories: reality (contemporary false… See the full description on the dataset page: https://huggingface.co/datasets/maxvonk/anti-factual-personas.

<iframe src="https://silytics.si/embed/item/ee54f9b3-2c0f-4c27-9585-24f0329ffece" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 18 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.