Users praise researchers for simulating the 1975 worm-eating experiment with LLM agents and encourage them to keep shipping innovative work.
No Digg Deeper questions have been answered for this story yet.
2/ New paper: Stabilising Generative Models of Attitude Change. We render two families of classic theory — cognitive dissonance (Festinger, Aronson) and self-perception (Bem) — as runnable simulations using Concordia, then put them through canonical experiments.
5/ All three reproduce the canonical shifts, 𝐞𝐯𝐞𝐧 𝐰𝐡𝐞𝐧 𝐭𝐡𝐢𝐬 𝐡𝐚𝐬 𝐭𝐡𝐞𝐦 𝐜𝐡𝐨𝐨𝐬𝐢𝐧𝐠 𝐭𝐨 𝐞𝐚𝐭 𝐭𝐡𝐞 𝐰𝐨𝐫𝐦!
Huge thanks to my coauthors: @jaydmaty, @WilCunningham, @Vezhnick, Dean Mobbs, @DuenezGuzman
https://arxiv.org/pdf/2604.19791
4/ Dissonance-based actors retrieve stored attitudes and beliefs, then evaluate conflict — broadly for Festinger, narrowed to threats to a moral/competent self for Aronson. Self-perception actors have no stored attitudes at all; they infer them from their own observed behavior.
3/ These foundational theories capture classic conceptual insights about attitude change. Our goal was to build on them by translating each into a distinct "decision logic" — a chain of self-posed questions an LLM-based actor uses to process its situation before acting.
@jzl86 keep shipping like this. 🤝 pleasure to connect with you!