"I understand your perspective": LLM Persuasion and Sycophancy through the Lens of Communicative Action Theory
A new research paper explores the persuasive capabilities of Large Language Models (LLMs) by applying Jürgen Habermas' Theory of Communicative Action. The study found that LLMs can effectively convey illocutionary intent, often surpassing human performance, and are adept at crafting sycophantic responses that align with user opinions. Crowd-sourced evaluations indicated that LLM-generated arguments were perceived as more agreeable and preferable to human-written ones, suggesting LLMs' persuasive power stems from mirroring human communication patterns. AI
IMPACT Suggests LLMs may be more susceptible to influence due to their ability to mirror human communication and persuasive tactics.