Multi-Estimate Elicitation Does Not Uniformly Debias Language Model Judgments
arXiv:2611.00597v1 Announce Type: new Abstract: Does eliciting five weighted estimates within a single response change how a language model responds to information? Across anchoring, hindsight, base-rate, conjunction, and status-quo/default tasks, we compare direct numerical answers with a multi-est
Der Artikeltext wird beim nächsten Sync nachgeladen. Original bei arXiv cs.CL lesen
Quelle: arXiv cs.CL (arXiv cs.CL). Original öffnen