Research map / Large language models / Language, safety and interpretability
Enhancing Uncertainty Estimation in LLMs with Expectation of Aggregated Internal Belief
Zeguan Xiao, Diyang Dou, Boya Xiong, Yun Chen, Guanhua Chen
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 |
|---|---|---|---|
| aggregate hidden states from the last n layers instead of only the last layer (ablation) | EAGLE (Llama3 8B) · TriviaQA | ECE x100: 18.8 -> 1.7 | Improved |
| aggregate hidden states from the last n layers instead of only the last layer (ablation) | EAGLE (Qwen2.5 7B) · GSM8K | ECE x100: 8.5 -> 3.0 | Improved |
| aggregate logits across layers rather than probabilities (ablation) | EAGLE (Llama3 8B) · TriviaQA | ECE x100: 63.5 -> 1.7 | Improved |
| use expectation over score distribution instead of the max-probability score (ablation) | EAGLE (Llama3 8B) · TriviaQA | ECE x100: 28.3 -> 1.7 | Improved |
| use expectation over score distribution instead of the max-probability score (ablation) | EAGLE (Qwen2.5 7B) · TriviaQA | ECE x100: 31.8 -> 22.1 | Improved |
| use EAGLE confidence to pick among 5 sampled answers | Llama3 8B · TriviaQA | accuracy: 71.3 -> 74.8 | Improved |
| use EAGLE confidence to pick among 5 sampled answers | Qwen2.5 7B · TriviaQA | accuracy: 67.5 -> 69.4 | Improved |
| replace the score-criteria prompt with a CoT explanation prompt (ablation) | EAGLE (Llama3 8B) · TriviaQA | ECE x100: 1.7 -> 13.6 | Hurt |
| remove score criteria from the self-evaluation prompt (ablation) | EAGLE (Llama3 8B) · TriviaQA | ECE x100: 1.7 -> 4.4 | Hurt |
| vary prompt variant (ablation) | EAGLE (Llama3 8B) · GSM8K | ECE x100: 7.6 / 6.6 / 5.8 | No clear effect |
| narrow the confidence score range from 0-9 to 0-1 (ablation) | EAGLE (Llama3 8B) · GSM8K | ECE x100: 7.6 -> 11.1 | Hurt |
| narrow the confidence score range from 0-9 to 0-4 (ablation) | EAGLE (Llama3 8B) · GSM8K | AUROC x100: 68.3 -> 64.7 | Hurt |
Closest papers by meaning
- Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs (ICLR 2024 · similarity 0.91)
- Large Language Models Must Be Taught to Know What They Don’t Know (NeurIPS 2024 · similarity 0.90)
- Benchmarking LLMs via Uncertainty Quantification (NeurIPS 2024 · similarity 0.89)
- An Empirical Analysis of Uncertainty in Large Language Model Evaluations (ICLR 2025 · similarity 0.89)
- Do LLMs estimate uncertainty well in instruction-following? (ICLR 2025 · similarity 0.89)
- Scalable Best-of-N Selection for Large Language Models via Self-Certainty (NeurIPS 2025 · 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.
