Pith. sign in

REVIEW 15 cited by

Fine-tuning Language Models for Factuality

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

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 2311.08401 v1 pith:ZRIZNLTI submitted 2023-11-14 cs.CL cs.AIcs.LG

Katherine Tian , Eric Mitchell , Huaxiu Yao , Christopher D. Manning , Chelsea Finn This is my paper · ORCID
classification cs.CLcs.AIcs.LG
keywords factualitylanguagemodelsgeneratedmodelpreferenceclaimscompared
verification ladder T0 review T1 audit T2 compute T3 formal

Signed reviews

No signed human review yet.

0 comments
read the original abstract

The fluency and creativity of large pre-trained language models (LLMs) have led to their widespread use, sometimes even as a replacement for traditional search engines. Yet language models are prone to making convincing but factually inaccurate claims, often referred to as 'hallucinations.' These errors can inadvertently spread misinformation or harmfully perpetuate misconceptions. Further, manual fact-checking of model responses is a time-consuming process, making human factuality labels expensive to acquire. In this work, we fine-tune language models to be more factual, without human labeling and targeting more open-ended generation settings than past work. We leverage two key recent innovations in NLP to do so. First, several recent works have proposed methods for judging the factuality of open-ended text by measuring consistency with an external knowledge base or simply a large model's confidence scores. Second, the direct preference optimization algorithm enables straightforward fine-tuning of language models on objectives other than supervised imitation, using a preference ranking over possible model responses. We show that learning from automatically generated factuality preference rankings, generated either through existing retrieval systems or our novel retrieval-free approach, significantly improves the factuality (percent of generated claims that are correct) of Llama-2 on held-out topics compared with RLHF or decoding strategies targeted at factuality. At 7B scale, compared to Llama-2-chat, we observe 58% and 40% reduction in factual error rate when generating biographies and answering medical questions, respectively.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 15 Pith papers

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

  1. From Evidence to Belief: A Bayesian Epistemology Approach to Language Models

    cs.AI 2025-04 conditional novelty 6.0 of 10

    Language models follow Bayesian confirmation with golden evidence but fail to follow Bayesian disconfirmation and irrelevance assumptions, and they are biased toward golden evidence.

  2. TruthFlow: Truthful LLM Generation via Representation Flow Correction

    cs.CL 2025-02 conditional novelty 6.0 of 10

    TruthFlow uses flow matching to produce query-specific representation corrections, improving truthfulness on TruthfulQA open-ended generation across several LLMs.

  3. Language Models Prefer What They Know: Relative Confidence Estimation via Confidence Preferences

    cs.CL 2025-02 conditional novelty 6.0 of 10

    Relative pairwise confidence comparisons aggregated by rank aggregation produce more reliable confidence scores for language models than direct absolute confidence prompts.

  4. Context-DPO: Aligning Language Models for Context-Faithfulness

    cs.CL 2024-12 conditional novelty 6.0 of 10

    Context-DPO fine-tunes LLMs with direct preference optimization on counterfactual passages, yielding 35-280% context-faithfulness gains on its new ConFiQA benchmark.

  5. ROSE: A Reward-Oriented Data Selection Framework for LLM Task-Specific Instruction Tuning

    cs.LG 2024-12 conditional novelty 6.0 of 10

    A reward-oriented influence scoring method that uses pairwise preference loss to select 5% of instruction-tuning data, outperforming similarity-based selection baselines on SHP, SE, and HH-RLHF.

  6. Quantized Delta Weight Is Safety Keeper

    cs.CR 2024-11 conditional novelty 6.0 of 10

    One-bit quantization of fine-tuning delta weights suppresses alignment-breaking and backdoor attacks on LLMs, at the price of measurable utility loss.

  7. EvolveSearch: An Iterative Self-Evolving Search Agent

    cs.CL 2025-05 conditional novelty 5.0 of 10

    An iterative loop of RL and filtered SFT on the agent's own rollouts improves a 7B web-search agent by a few accuracy points on multi-hop QA benchmarks.

  8. Improve Decoding Factuality by Token-wise Cross Layer Entropy of Large Language Models

    cs.CL 2025-02 conditional novelty 5.0 of 10

    A new decoding-time method, END, uses per-token cross-layer entropy of prediction growth to boost factual tokens, improving truthfulness and informativeness on hallucination benchmarks.

  9. Aligning Large Language Models for Faithful Integrity Against Opposing Argument

    cs.CL 2025-01 conditional novelty 5.0 of 10

    An LLM is fine-tuned with DPO to make the strength of its stance in conversation match its self-estimated confidence, improving resistance to misleading arguments and receptiveness to corrections.

  10. Beyond Logit Lens: Contextual Embeddings for Robust Hallucination Detection & Grounding in VLMs

    cs.CL 2024-11 conditional novelty 5.0 of 10

    Middle-layer contextual embeddings, not logit-lens readings, improve hallucination detection in VLMs and enable bounding-box grounding for visual question answering.

  11. Efficient Alignment of Large Language Models via Data Sampling

    cs.LG 2024-11 reject novelty 5.0 of 10

    Alignment performance plateaus quickly with data, and a GMM-plus-entropy sampling method (ISA) can allegedly match full-dataset alignment with under 10% of the data, though the supporting evidence is weak.

  12. Inteligencia Artificial jur\'idica y el desaf\'io de la veracidad: an\'alisis de alucinaciones, optimizaci\'on de RAG y principios para una integraci\'on responsable

    cs.AI 2025-09 conditional novelty 4.0 of 10

    Legal AI hallucination persists in commercial RAG tools (17-34%+ of queries), so the report argues the fix is consultative, source-citing system design plus mandatory human oversight, not better generative models.

  13. An Integrated Framework of Prompt Engineering and Multidimensional Knowledge Graphs for Legal Dispute Analysis

    cs.AI 2025-07 conditional novelty 4.0 of 10

    A prompt-plus-knowledge-graph framework for legal dispute analysis reports improved LLM sensitivity and citation accuracy on a 100-pair test set, but with limited statistical support.

  14. MPPO: Multi Pair-wise Preference Optimization for LLMs with Arbitrary Negative Samples

    cs.CL 2024-12 conditional novelty 4.0 of 10

    MPPO trains LLMs without a reference model by treating the geometric mean of response-token probabilities as the reward and jointly suppressing multiple negative responses; the Pair-MNM variant reports the best MT-Ben...

  15. Building Task Bots with Self-learning for Enhanced Adaptability, Extensibility, and Factuality

    cs.CL 2025-08 conditional novelty 2.0 of 10

    A thesis that combines self-learning from dialog logs, schema-guided prompting, and self-aligned factuality to build task bots with minimal human intervention.

Pith tools