Research map / Large language models / Language, safety and interpretability
Semantic Volume: Quantifying and Detecting Both External and Internal Uncertainty in LLMs
Xiaomin Li, Zhou Yu, Ziji Zhang, Yingying Zhuang, Swair Shah, Narayanan Sadagopan, Anurag Beniwal
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 |
|---|---|---|---|
| add PCA dimension reduction to the embeddings (ablation) | Semantic Volume · CLAMBER and TriviaQA | F1 | Improved |
| increase number of perturbations n (ablation) | Semantic Volume · CLAMBER and TriviaQA | F1 | Improved |
| use larger embedding model (Qwen2-7B instead of Qwen2-1.5B) (ablation) | Semantic Volume · CLAMBER | accuracy / F1: 58.0 -> 58.1 accuracy; 69.1 -> 68.0 F1 | No clear effect |
| use NV-Embed instead of Qwen2-1.5B embeddings (ablation) | Semantic Volume · CLAMBER | accuracy / F1: 58.0 -> 58.4 accuracy; 69.1 -> 67.9 F1 | No clear effect |
| replace semantic entropy clustering with Gram-determinant volume | uncertainty measure · CLAMBER | F1: 62.8 -> 69.1 | Improved |
| replace semantic entropy with Semantic Volume | uncertainty measure on Llama3.2-1B responses · TriviaQA | AUROC: 73.9 -> 79.7 | Improved |
| replace Lexical Similarity with Semantic Volume | uncertainty measure on Llama3.2-1B responses · SQuAD | F1: 59.9 -> 71.2 | Improved |
| replace Last Token Entropy with Semantic Volume | uncertainty measure · AmbigQA | F1: 63.2 -> 67.6 | Improved |
| use larger response generator (Llama3-8B instead of Llama3.2-1B) | most detection methods · TriviaQA / SQuAD | detection performance | Mixed |
Closest papers by meaning
- Improving Uncertainty Estimation through Semantically Diverse Language Generation (ICLR 2025 · similarity 0.92)
- INSIDE: LLMs' Internal States Retain the Power of Hallucination Detection (ICLR 2024 · similarity 0.92)
- LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinations (ICLR 2025 · similarity 0.91)
- Enhancing Uncertainty Modeling with Semantic Graph for Hallucination Detection (AAAI 2025 · similarity 0.91)
- Kernel Language Entropy: Fine-grained Uncertainty Quantification for LLMs from Semantic Similarities (NeurIPS 2024 · similarity 0.91)
- In-Context Sharpness as Alerts: An Inner Representation Perspective for Hallucination Mitigation (ICML 2024 · similarity 0.90)
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.
