Reproduce “Language-model agents' beliefs can be manipulated via truthful evidence in open channel” 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

Jinwei Hu, Qisong He, Xinmiao Huang, Youcheng Sun, Yi Dong, Xiaowei Huang

The authors test Generative Montage, in which coordinated agents select and arrange truthful posts to invite unsupported inferences without stating a rumour target. In simulated feeds, 12 of 14 models accepted unsupported targets in more than 65% of runs. Grounded Monitor’s warning-plus-uncertainty condition improved accuracy, but checks only for direct support in a single post, not evidence combined across posts.

Summary by alphaXiv.

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