Research map / Large language models / Language, safety and interpretability
Estimating Semantic Alphabet Size for LLM Uncertainty Quantification
Lucas Hurley McCabe, Rimon Melamed, Thomas Hartvigsen, H Howie Huang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| adjust DSE for sample coverage using hybrid alphabet size estimator | discrete semantic entropy · HotpotQA, SQuAD 2.0, POTATO, BioASQ (n=10) | MSE vs SE*: BioASQ Gemma-2: 1.64 -> 0.78 | Improved |
| use hybrid instead of Good-Turing Chao-Shen coverage adjustment | DSE · SQuAD 2.0 Mistral-v0.3 | MSE vs SE*: 0.86 -> 0.68 | Improved |
| use hybrid estimator on Gemma-3 (12B) | DSE · POTATO | entropy bias: slightly overshoots SE* | Mixed |
| use hybrid DSE rather than other explicit SE estimators for incorrectness detection | DSE · HotpotQA, SQuAD, BioASQ, POTATO | AUROC rank (Bradley-Terry): highest rank among explicit SE estimators | Improved |
| use semantic alphabet size estimators (Hybrid, U_EigV) alone | LLM uncertainty · incorrectness detection | Bradley-Terry rank: rank CI [1, 4] and [1, 3] | Improved |
| use white-box SE (response probabilities) vs plugin DSE | SE · incorrectness detection | AUROC: qualitatively similar | No clear effect |
Closest papers by meaning
- Kernel Language Entropy: Fine-grained Uncertainty Quantification for LLMs from Semantic Similarities (NeurIPS 2024 · similarity 0.92)
- Improving Uncertainty Estimation through Semantically Diverse Language Generation (ICLR 2025 · similarity 0.89)
- Benchmarking LLMs via Uncertainty Quantification (NeurIPS 2024 · similarity 0.89)
- Inv-Entropy: A Fully Probabilistic Framework for Uncertainty Quantification in Language Models (NeurIPS 2025 · similarity 0.89)
- Semantic Density: Uncertainty Quantification for Large Language Models through Confidence Measurement in Semantic Space (NeurIPS 2024 · similarity 0.89)
- Uncertainty Quantification for Large Language Diffusion Models (NeurIPS 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
