Study Finds Multi-Estimate Elicitation Does Not Uniformly Debias Language Model Judgments
A new arXiv paper examines whether eliciting five weighted estimates within a single response changes how language models respond to information, testing anchoring, hindsight, and base-rate biases.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-11-03 05:00 · arXiv cs.CL
Multi-Estimate Elicitation Does Not Uniformly Debias Language Model Judgments