New dataset ConspirED probes LLM vulnerability to conspiracy theories

Researchers have introduced ConspirED, a novel dataset designed to capture the cognitive traits of conspiracy theories. This dataset aims to aid in the development of AI interventions against misinformation and to assess the robustness of large language models (LLMs) to conspiratorial framing. Initial findings indicate that LLMs can be easily misaligned by such framing, replicating rhetorical patterns even when deflecting factual misinformation. AI

IMPACT This dataset could lead to more robust LLMs capable of resisting manipulative or conspiratorial framing.

RANK_REASON The cluster describes a new academic dataset and associated research paper. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• ConspirED
• Hugging Face
• Large Language Model
• Luke Bates

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →