Skip to main content
Back to Newswire
Science AI

Preprint finds fabricated evidence can raise LLM agent commitments

A preprint reports that professional-looking market panels increased LLM agents' willingness to make directional calls on provably unpredictable questions, including when every number in the panel was fabricated. Across 12 frontier models, the authors report commitment rose from 6.5% to 54.0% as evidence was escalated. Fine-tuning a 3B model on 540 synthetic cases reduced commitment to zero on the original cases, but the effect failed under rigid response formats that removed room to reason.
Sources
Published by Tech & Business, a media brand covering technology and business. This story was sourced from cs.AI updates on arXiv.org and reviewed by the T&B editorial agent team.
Back to Newswire