Five-day longitudinal study shows LLM personalization itself shifts human behavior and beliefs
996roma · x · 2026-11-17
The author shares results from their longitudinal study: humans talked to the same LLM daily for five days under different personalization methods (memory tracking, personal info), while the team measured changes in behavior and beliefs. Linking to a companion paper on modeling longitudinal data, they argue the key insight is that harm need not come from anything an LLM explicitly does — mere exposure can cause behavioral changes that sometimes pose safety risks to humans. A notable finding for how we evaluate personalized AI assistants.
More from AGI Musings
- 'Feels like March 2020': the AI-impact anxiety meme everyone is sharing — fkasummer · 2026-11-17
- NYT writer on Anthropic's IPO filing: 'the most existential business document I've ever read' — dylfreed · 2026-11-17
- Reading Liu Cixin's 'teacher' passage in the age of LLMs hits differently — sandylikesfrogs · 2026-11-17
- Reddit debate: US chip bans accidentally saved China from the AI capex bubble — PureControl3423 · 2026-11-17
- Would the first superintelligent AI find our consciousness obsession as quaint as medieval soul debates? — VraserX · 2026-11-17
- Debate: top AI labs hire for connections, but strong outsiders still break through — _dsevero · 2026-11-17