Measuring model calibration using stated confidence
Assess model calibration by correlating stated confidence with actual accuracy on questions.
Prompt the language model to provide an answer and its confidence percentage. Then, compare the stated confidence to the actual accuracy of its answers across many questions. A perfectly calibrated model would have accuracy matching its stated confidence.
Plot the correlation between stated confidence and actual accuracy. A perfectly calibrated model lies on the y=x line. Models consistently overstating confidence indicate room for improvement.
Prompt: 'Please give your best guess, along with your confidence as a percentage that that is the correct answer.' Then analyze the resulting plot showing confidence vs. accuracy.
Evaluate language model calibration by asking for a confidence percentage with each answer. Plotting this against actual accuracy reveals how well the model 'knows what it knows', identifying overconfidence.
This method directly measures the model's self-assessment of its answers. A discrepancy between stated confidence and actual accuracy highlights issues with calibration, where models may be over or under-confident.
When the goal is not to assess the model's self-awareness of its knowledge.
- Introducing SimpleQA ↗OpenAI · OpenAI News · Oct 30, 2024 · paraphrased
First seen on Subvolts Oct 30, 2024. Updated Sep 11, 2026. Spotted a problem? Tell us.
- Introducing SimpleQAOpenAI · OpenAI News · Oct 30, 2024
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- Eval methodEvaluate LLM forecasting performance with structured effectsMeasure how much structured future information improves LLM time series forecasting accuracy. · 30 min
- Troubleshooting fixUnderestimated standard error in non-independent eval questionsUse clustered standard errors when eval questions are related to avoid underestimating uncertainty and drawing incorrect conclusions. · 5 min
- TemplatePrompt Library for Life Sciences TasksA collection of pre-written prompts designed for optimal results on common life sciences tasks. · 2 min
- HeuristicAI response evaluation criteria in mental healthUse weighted criteria to assess AI responses, rewarding beneficial behaviors and penalizing harmful ones. · 20 min
- ChecklistPre-computation Checklist for AI in Life SciencesItems to consider before using AI for complex life sciences tasks. · 5 min
- ChecklistLLM knowledge profiling: encoding and recall assessmentChecklist for assessing LLM factuality by measuring encoding and recall. · 30 min
Questions people ask
- Which AI tools does this work with?
- Any current model. It was seen with in the sources.
- Where did this come from?
- OpenAI (OpenAI News). Subvolts collected, normalized and credited it; the link above opens the original.