Reproduce “Preference Optimization Drives Monoculture in LLM Prediction Markets” 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
James Begin, Brendan Gho, Suman Muppavarapu, Tyson Tsay, Atharva Mohan, Afnan Shaik, Ruizhe Li, Vasu Sharma and 1 more
The authors test whether same-model agents improve simulated prediction markets and whether preference optimization increases their shared errors. In TruthfulQA binary questions, adding Llama 3.1 8B agents produced no observed accuracy gain as the group grew. Tulu 3 8B error correlation rose from 0.559 for SFT to 0.799 for DPO, supporting a checkpoint-specific effect, not a universal claim.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.