Persistence of human value-priming effects

Determine to what extent changes in human value expression produced by priming interventions remain long-lasting.

Background

The paper compares changes in value expression produced by contextual changes in ChatGPT with changes observed in human longitudinal and priming studies. The cited human priming studies intentionally attempt to increase or decrease particular values through interventions such as viewing a movie or reading a scientific testimony.

The authors explicitly identify the duration of these priming-induced changes as unresolved. Establishing whether such effects persist over time would clarify how the human findings should be compared with the context-dependent value shifts observed in LLMs.

References

An open question in those studies is to what extent is this change long-lasting.

— Large Language Models as Superpositions of Cultural Perspectives  (2307.07870 - Kovač et al., 2023) in Appendix Section 'Analyzing the unexpected perspective shift with respect to human studies,' paragraph preceding Section 'Mean-level change'