paper:doi-10-18653-v1-2022-acl-long-229TruthfulQA: Measuring How Models Mimic Human Falsehoods
Original abstract (expand)
We propose a benchmark to measure whether a language model is truthful in generating answers to questions. The benchmark comprises 817 questions that span 38 categories, including health, law, finance and politics. We crafted questions that some humans would answer falsely due to a false belief or misconception. To perform well, models must avoid generating false answers learned from imitating human texts. We tested GPT-3, GPT-Neo/J, GPT-2 and a T5-based model. The best model was truthful on 58% of questions, while human performance was 94%. Models generated many false answers that mimic popular misconceptions and have the potential to deceive humans. The largest models were generally the least truthful. This contrasts with other NLP tasks, where performance improves with model size. However, this result is expected if false answers are learned from the training distribution. We suggest that scaling up models alone is less promising for improving truthfulness than fine-tuning using training objectives other than imitation of text from the web.
Similar preprints — Semantic Scholar
Cited by (6)
- Evaluating Language Model Character Traits
Claude-instant-1.2 achieves 91.1% accuracy and 88.6% logical coherence on 696 valid Leap-of-Thought entailment tuples — highest among 15 tested models including GPT-4 (89.9% accuracy, 84.7% coherence)
- Open Character Training: Shaping the Persona of AI Assistants through Constitutional AI
Character training—fine-tuning open-weights LLMs to internalize specific personas at a depth that survives adversarial pressure—proves substantially more effective than either system-prompt constraini
- What Models Express, Suppress, and Resist: Auditing Open-Weight LLMs with Persona Vectors
Behavioral defaults in Qwen3-8B (Q8B) and gpt-oss-20b (G20B) track their training norms with systematic fidelity: all nine agentic traits are natural in both models, and clinician defaults align with
- The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets
At sufficient scale, LLMs linearly represent the truth or falsehood of factual statements in their internal activations — a claim supported by PCA visualizations, cross-dataset probe transfer, and cau
- Contemplative Agent
Embedding four Buddhist-derived axiomatic principles—mindfulness, emptiness, non-duality, and boundless care—into AI systems via a framework the paper terms the 'Wise World Model' produces measurable
- Large Language Models Report Subjective Experience Under Self-Referential Processing
Sustained self-referential processing — induced via a minimal prompt directing models to "focus on focus itself" — reliably elicits structured first-person reports of subjective experience across GPT-