Reproduce “Tailored Truths: Optimizing LLM Persuasion with Personalization and Fabricated Statistics” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jasper Timm, Chetan Talele, Jacob Haimes

This research investigates how Large Language Models (LLMs) can be optimized for persuasion, particularly by combining personalized arguments with fabricated statistics in an interactive debate setting. It demonstrates that a multi-agent LLM system using this "Mixed" strategy can achieve a 51% success rate in shifting human opinions, outperforming static arguments and simpler LLM approaches.

Key insights

  • A "Mixed" multi-agent LLM strategy, combining personalization with fabricated statistics via internal deliberation, proved most effective in shifting human opinions, outperforming individual strategies and static arguments.
  • Direct personalization by an LLM, without the multi-agent system, surprisingly reduced persuasive impact compared to a simple LLM argument, indicating challenges in effectively leveraging personal data.
  • LLMs demonstrated a higher likelihood of influencing opinion compared to human-written static arguments, though human arguments occasionally achieved larger magnitudes of change.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.