English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

The increased deployment of LMs for real-world tasks involving knowledge andfacts makes it important to understand model epistemology: what LMs think theyknow, and how their attitudes toward that knowledge are affected by languageuse in their inputs. Here, we study an aspect of model epistemology: howepistemic markers of certainty, uncertainty, or evidentiality like "I'm sureit's", "I think it's", or "Wikipedia says it's" affect models, and whether theycontribute to model failures. We develop a typology of epistemic markers andinject 50 markers into prompts for question answering. We find that LMs arehighly sensitive to epistemic markers in prompts, with accuracies varying morethan 80%. Surprisingly, we find that expressions of high certainty result in a7% decrease in accuracy as compared to low certainty expressions; similarly,factive verbs hurt performance, while evidentials benefit performance. Ouranalysis of a popular pretraining dataset shows that these markers ofuncertainty are associated with answers on question-answering websites, whilemarkers of certainty are associated with questions. These associations maysuggest that the behavior of LMs is based on mimicking observed language use,rather than truly reflecting epistemic uncertainty.

Navigating the Grey Area: How Expressions of Uncertainty and Overconfidence Affect Language Models