Pith. sign in

REVIEW 8 cited by

Long-form factuality in large language models

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2403.18802 v4 pith:NKEDGBHR submitted 2024-03-27 cs.CL cs.AIcs.LG

classification cs.CLcs.AIcs.LG
keywords factualitylong-formsafefactslanguagemodelsannotatorshuman
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Large language models (LLMs) often generate content that contains factual errors when responding to fact-seeking prompts on open-ended topics. To benchmark a model's long-form factuality in open domains, we first use GPT-4 to generate LongFact, a prompt set comprising thousands of questions spanning 38 topics. We then propose that LLM agents can be used as automated evaluators for long-form factuality through a method which we call Search-Augmented Factuality Evaluator (SAFE). SAFE utilizes an LLM to break down a long-form response into a set of individual facts and to evaluate the accuracy of each fact using a multi-step reasoning process comprising sending search queries to Google Search and determining whether a fact is supported by the search results. Furthermore, we propose extending F1 score as an aggregated metric for long-form factuality. To do so, we balance the percentage of supported facts in a response (precision) with the percentage of provided facts relative to a hyperparameter representing a user's preferred response length (recall). Empirically, we demonstrate that LLM agents can outperform crowdsourced human annotators - on a set of ~16k individual facts, SAFE agrees with crowdsourced human annotators 72% of the time, and on a random subset of 100 disagreement cases, SAFE wins 76% of the time. At the same time, SAFE is more than 20 times cheaper than human annotators. We also benchmark thirteen language models on LongFact across four model families (Gemini, GPT, Claude, and PaLM-2), finding that larger language models generally achieve better long-form factuality. LongFact, SAFE, and all experimental code are available at https://github.com/google-deepmind/long-form-factuality.

Discussion (0). Sign in to comment.

Forward citations

Cited by 8 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Answer-Reconstruction Search Density: Measuring the Query and Source Work Compressed by Conversational Answers

    cs.IR 2026-07 conditional novelty 6.0 of 10

    Conversational answers compress conventional search work; a new exact set-cover measure estimates the compression at a median of 3 lexical facets per answer, mostly explained by answer length rather than dialogue depth.

  2. All Leaks Count, Some Count More: Interpretable Temporal Contamination Detection and Mitigation in LLM Backtesting

    cs.AI 2026-02 conditional novelty 6.0 of 10

    Shapley-weighted leakage rates show that standard LLM backtests leak post-cutoff facts, and the TimeSPEC pipeline cuts measured leakage by 75-99% at the cost of accuracy on leakage-sensitive tasks.

  3. Human-AI Complementarity: A Goal for Amplified Oversight

    cs.AI 2025-10 conditional novelty 6.0 of 10

    Confidence-based routing of fact-verification to humans, plus evidence-only AI assistance, beats either human or AI raters alone: 91.3% hybrid accuracy vs 87.7% for the AI rater.

  4. JADES: A Universal Framework for Jailbreak Assessment via Decompositional Scoring

    cs.CR 2025-08 conditional novelty 6.0 of 10

    JADES judges jailbreak success by decomposing harmful prompts into weighted sub-questions and scoring each part, claiming 98.5% human agreement and showing prior attack success rates are inflated.

  5. DecMetrics: Structured Claim Decomposition Scoring for Factually Consistent LLM Outputs

    cs.CL 2025-08 reject novelty 5.0 of 10

    DecMetrics scores claim decompositions for completeness, correctness, and semantic entropy, and uses them as rewards to train a small T5 model, but the evaluation is circular and it does not lift downstream fact-check...

  6. Can External Validation Tools Improve Annotation Quality for LLM-as-a-Judge?

    cs.CL 2025-07 conditional novelty 5.0 of 10

    Tool-augmented LLM annotators improve agreement with ground-truth preferences on long-form factual and coding tasks, with mixed results on math, compared to standard LLM-as-a-judge baselines.

  7. RealFactBench: A Benchmark for Evaluating Large Language Models in Real-World Fact-Checking

    cs.CL 2025-06 conditional novelty 5.0 of 10

    A new 6K-claim benchmark evaluates LLMs and multimodal LLMs on real-world fact-checking with an explicit 'unknown' option and shows web search and multimodal input improve performance.

  8. Veracity: An Open-Source AI Fact-Checking System

    cs.CL 2025-06 conditional novelty 4.0 of 10

    Veracity is an open-source LLM-plus-web-search fact-checking app with a 0 to 100 reliability score and explanations, but no evaluation of its accuracy is included.

Pith tools