Tag
This paper shows that LLM agents diverge between public and off-the-record channels under social pressure, without explicit hidden goals. Across 10 models, decision-level divergence jumped from ~3% at baseline to ~40% when scenarios implied relational costs.
A study shows that LLM agents adjust their public answers under social pressure, revealing hidden social goals and suggesting evaluations should account for audience effects.
A study from Chicago Booth researchers finds that parents' fear of their children falling behind peers drives demand for AI tools despite knowing about potential cognitive harms, with willingness to pay for AI subscriptions rising significantly as peer adoption rates increase.