{"slug": "spanuq-span-level-uncertainty-quantification-for-large-language-model-generation", "title": "SpanUQ: Span-Level Uncertainty Quantification for Large Language Model Generation", "summary": "Researchers introduced SPANUQ, a lightweight probe for span-level uncertainty quantification in large language models, achieving 10-20x faster inference than sampling-based methods while outperforming them in uncertainty quality. The system uses a DETR-style decoder to detect semantically coherent spans and estimate their uncertainty, enabling precise error localization that sequence-level methods cannot provide.", "body_md": "arXiv:2607.05721v1 Announce Type: new\nAbstract: Uncertainty estimation is essential not only for the trustworthy deployment of large language models (LLMs) but also as a foundation for self-refinement in LLM generation. However, existing approaches operate at suboptimal granularities: token-level scores lack semantic coherence, while sequence-level scores fail to localize errors. We formalize Span-Level Uncertainty Estimation (SLUE), a new task that targets the natural granularity for uncertainty: semantically coherent text spans, each conveying a single assessable unit of meaning. To address this task, we introduce SPANUQ, a lightweight probe that distills the uncertainty knowledge from expensive multi-sample inference into a single forward pass over LLM hidden states. SPANUQ employs a DETR-style span decoder to simultaneously detect spans and estimate their uncertainty via a Mixture of Beta distribution, trained with a principled combination of Beta NLL regression and contrastive ranking objectives. We construct SPANUQ-BENCH, the first span-level uncertainty benchmark comprising 20K prompts, 293K annotated spans, and continuous soft labels derived from multi-sample claim verification. Experiments on five LLM backbones show that SPANUQ consistently achieves the best span-level uncertainty quality, outperforming the strongest probe baseline and all sampling-based methods while being 10-20x faster. Its DETR-based span detector attains 0.910 F1, surpassing the best heuristic by 39.4%, enabling precise error localization that sequence-level methods cannot provide. The framework generalizes across five LLMs spanning two model families.", "url": "https://wpnews.pro/news/spanuq-span-level-uncertainty-quantification-for-large-language-model-generation", "canonical_source": "https://arxiv.org/abs/2607.05721", "published_at": "2026-07-08 04:00:00+00:00", "updated_at": "2026-07-08 04:03:49.188603+00:00", "lang": "en", "topics": ["large-language-models", "ai-research", "machine-learning", "natural-language-processing", "ai-safety"], "entities": ["SPANUQ", "DETR", "SLUE", "SPANUQ-BENCH", "arXiv"], "alternates": {"html": "https://wpnews.pro/news/spanuq-span-level-uncertainty-quantification-for-large-language-model-generation", "markdown": "https://wpnews.pro/news/spanuq-span-level-uncertainty-quantification-for-large-language-model-generation.md", "text": "https://wpnews.pro/news/spanuq-span-level-uncertainty-quantification-for-large-language-model-generation.txt", "jsonld": "https://wpnews.pro/news/spanuq-span-level-uncertainty-quantification-for-large-language-model-generation.jsonld"}}