Challenge: Large language models (LLMs) are trained on vast corpora that contain substantial knowledge but their outputs often contain confidently stated inaccuracies.
Approach: They propose to encode truthfulness as a distinct linear feature, termed the "truth direction", which can classify truthfulness reliably.
Outcome: The proposed model can generalize to logical transformations, question-answering tasks, in-context learning, and external knowledge sources.

Similar Papers

On the Universal Truthfulness Hyperplane Inside LLMs (2024.emnlp-main)

Copied to clipboard

Challenge: Recent studies have explored hallucinations through the lens of internal representations, proposing mechanisms to decipher LLMs’ adherence to facts.
Approach: They propose to train a universal truthfulness hyperplane that distinguishes the model’s factually correct and incorrect outputs on a diverse collection of over 40 datasets and examine its cross-task, cross-domain, and in-domain generalization.
Outcome: The proposed model is able to distinguish factual outputs from incorrect outputs on a diverse collection of over 40 datasets.
LLM Beliefs Are in Their Heads (2026.acl-long)

Copied to clipboard

Challenge: Using linear controlled probes, we investigate belief-like representations in decoder-only autoregressive LLMs using residual stream activations and single attention heads.
Approach: They develop four different experiments on decoder-only autoregressive LLMs and examine how they fare against these standards.
Outcome: The proposed representations exhibit strong truth sensitivity and consistent accuracy across models and data sets.
Quantized but Deceptive? A Multi-Dimensional Truthfulness Evaluation of Quantized LLMs (2025.emnlp-main)

Copied to clipboard

Challenge: Quantization enables efficient deployment of large language models in resource-constrained environments . but impact on truthfulness remains largely unexplored .
Approach: They propose a framework to assess the truthfulness of quantized large language models . they find quantized models retain internally truthful representations but produce false outputs .
Outcome: The framework assesses the truthfulness of quantized models across three dimensions . it finds that quantized model models retain internally truthful representations but are more susceptible to false outputs .
AI-LieDar : Examine the Trade-off Between Utility and Truthfulness in LLM Agents (2025.naacl-long)

Copied to clipboard

Challenge: LieDar is a framework to study how LLM-based agents navigate these scenarios in a multi-turn interactive setting.
Approach: They propose a framework to study how LLM-based agents navigate these scenarios in an interactive multi-turn setting.
Outcome: The proposed framework shows that all models are truthful less than 50% of the time, although truthfulness and goal achievement rates vary across models.
Are the Hidden States Hiding Something? Testing the Limits of Factuality-Encoding Capabilities in LLMs (2025.acl-long)

Copied to clipboard

Challenge: Recent studies suggest that LLMs encode internal representations of factuality when generating inaccurate or fabricated content.
Approach: They propose a strategy for sampling plausible true-false factoid sentences from tabular data and a procedure for generating realistic, LLM-dependent true-False datasets from Question Answering collections.
Outcome: The proposed approach lays the groundwork for future research on factuality in LLMs and offers practical guidelines for more effective evaluation.
How Context Shapes Truth: Geometric Transformations of Statement-level Truth Representations in LLMs (2026.acl-long)

Copied to clipboard

Challenge: Prior work shows that large language models encode whether a statement is true as a vector in residual stream activations.
Approach: They study how truth vectors change when context is introduced in Large Language Models . they measure directional change between truth vector with and without context and relative magnitude of truth vector upon adding context.
Outcome: The results show that large models distinguish relevant from irrelevant context mainly through directional change ()
Cognitive Dissonance: Why Do Language Model Outputs Disagree with Internal Representations of Truthfulness? (2023.emnlp-main)

Copied to clipboard

Challenge: Past work has found that query–probe disagreements occur in situations where either probes or queries are uncertain.
Approach: They propose to identify three different classes of disagreement, which they term confabulation, deception, and heterogeneity.
Outcome: The proposed methods are more accurate than LM outputs and better on uncertain answers than queries and probes.
Two Pathways to Truthfulness: On the Intrinsic Encoding of LLM Hallucinations (2026.acl-long)

Copied to clipboard

Challenge: Previous work shows that large language models generate hallucinations, yet the origins and mechanisms of these signals remain unclear.
Approach: They propose to validate and disentangle two different pathways for truthfulness cues . they also propose to use the same mechanism to derive self-contained evidence from the generated answer .
Outcome: The proposed applications improve hallucination detection performance by integrating two different inputs.
Truth Knows No Language: Evaluating Truthfulness Beyond English (2025.acl-long)

Copied to clipboard

Challenge: a new benchmark evaluates the truthfulness of large language models (LLMs) based on imitative falsehoods.
Approach: They propose a professionally translated extension of the TruthfulQA benchmark . it evaluates truthfulness in Basque, Catalan, Galician, and Spanish .
Outcome: The proposed extension of the TruthfulQA benchmark evaluates truthfulness in Basque, Catalan, Galician, and Spanish.
The Internal State of an LLM Knows When It’s Lying (2023.findings-emnlp)

Copied to clipboard

Challenge: Large Language Models (LLMs) have shown exceptional performance in various tasks, but one of their main drawbacks is generating inaccurate or false information with a confident tone.
Approach: They propose to train a classifier that outputs the probability that a statement is truthful based on the hidden layer activations of the LLM as it reads or generates the statement.
Outcome: The proposed classifier achieves an average of 71% to 83% accuracy labeling which sentences are true versus false, depending on the LLM base model.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations