Screenshot from GPT-4 about biases.
journal contribution
posted on 2025-04-28, 17:26 authored by Melanie Brucks, Olivier ToubiaThe following screenshot was taken on 03/08/2024. We use the API Playground to mimic the conditions of our experiment: GPT-4 pinned on 06/13/23, with a temperature of 0. The “System” portion of the prompt was set to its default (“You are a helpful assistant”).
(PDF)
History
Usage metrics
Categories
Keywords
term &# 8220seemingly arbitrary wayreduce methodological artifactsproduce methodological artifactsinfluences responses providedhidden associations embeddedfind robust evidencereduce either biasuncommon labels reducesprompt contains characteristicsmay subtly interactalleviate either responselarge language modelsfive largestatistical biasxlink ">using llmsseveral featuresprompt architectureobserve differencesllms demonstratellms ).inherent fallibilityindividual promptfindings highlightexacerbates response1 document1 ).
Licence
Exports
RefWorks
BibTeX
Ref. manager
Endnote
DataCite
NLM
DC