Researchers study opinion dynamics in LLM-LLM debates on climate change
A new arXiv preprint introduces a controlled framework where two language models debate a contested topic — climate-change attribution — to measure how opinions shift in extended conversations. One model receives an assigned stance and sometimes an explicit instruction to persuade; the other starts with no assigned position. The study tests four ordered pairings of GPT-6 Luna and Gemini 2.5…
Key points
- Framework tests LLM-LLM debates on climate-change attribution with assigned stances
- Same-model pairs narrow opinion gaps for one model; cross-model pairs widen gaps
- Explicit persuasion instruction does not reliably increase influence across conditions
Results show that same-model interactions tend to narrow the opinion gap, but only clearly for one of the two models. Cross-model interactions widen the gap. Explicit self-reports and in-dialogue behavior capture different aspects of opinion change, and a blanket instruction to persuade does not reliably increase influence. The authors argue that conversational influence depends on the specific models involved, not just assigned roles or initial positions, with implications for how opinion change is measured in LLM interactions.
Model page: GPT-6 Luna →
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- PHP patches critical RCE flaws discovered with GPT-5.6 Sol, affecting up to 70% of websites · 9 src
- Researchers release madartb, a parallel Multi-Dialect Arabic dependency treebank · 1 src
- Researchers decode 26 legal factors in three open LLMs · 1 src
- Researchers test how reinforcement learning teaches AI agents to enforce market collusion · 1 src
- LLMs can assess psychotherapy fidelity, study shows · 1 src
Comments
via GitHub Discussions