{"slug": "conformal-factuality-control-for-multi-hop-retrieval-augmented-generation", "title": "Conformal Factuality Control for Multi-Hop Retrieval-Augmented Generation", "summary": "Split-conformal claim filtering raises the share of multi-hop RAG responses whose retained claims are fully supported to 95.80%-97.20% at a 95% target, up from 55.60%-76.03% without filtering, according to an arXiv paper (2609.38222v1) evaluated on HotpotQA, Natural Questions and TriviaQA with Llama 3.1 8B and GPT-4o-mini. The gain is selective: only 4.41%-31.09% of generated claims are retained and 9.70%-51.40% of responses remain non-empty at that target, so the authors conclude nominal reliability must be read alongside claim retention and abstention.", "body_md": "arXiv:2609.38222v1 Announce Type: new \nAbstract: Retrieval-augmented generation (RAG) can ground large language models in external evidence, but retrieved context does not guarantee that generated claims are factually supported. This problem is especially relevant in multi-hop RAG, where retrieval and reasoning proceed through multiple dependent stages. We study whether claim-level conformal factuality control, previously developed for RAG, remains effective in this setting. We apply split-conformal claim filtering to multi-hop RAG and evaluate it on HotpotQA, Natural Questions, and TriviaQA using Llama 3.1 8B and GPT-4o-mini, together with a single-hop reference experiment. Across all six multi-hop model-dataset configurations, increasingly stringent conformal targets consistently increase the fraction of responses whose retained claims are fully supported. At the 95% target, this rate ranges from 95.80% to 97.20%, compared with 55.60%-76.03% without filtering. However, the improvement is strongly selective: only 4.41%-31.09% of generated claims are retained and 9.70%-51.40% of responses remain non-empty at the 95% target. These results show that conformal factuality extends to multi-hop RAG, while demonstrating that nominal reliability must be interpreted jointly with claim retention and abstention.", "url": "https://wpnews.pro/news/conformal-factuality-control-for-multi-hop-retrieval-augmented-generation", "canonical_source": "https://arxiv.org/abs/2609.38222", "published_at": "2026-10-01 04:00:00+00:00", "updated_at": "2026-10-01 04:19:04.367326+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-research", "natural-language-processing"], "entities": ["HotpotQA", "Natural Questions", "TriviaQA", "Llama 3.1 8B", "GPT-4o-mini", "arXiv"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/conformal-factuality-control-for-multi-hop-retrieval-augmented-generation", "markdown": "https://wpnews.pro/news/conformal-factuality-control-for-multi-hop-retrieval-augmented-generation.md", "text": "https://wpnews.pro/news/conformal-factuality-control-for-multi-hop-retrieval-augmented-generation.txt", "jsonld": "https://wpnews.pro/news/conformal-factuality-control-for-multi-hop-retrieval-augmented-generation.jsonld"}}