Reproduce “RLPF: Reinforcement Learning from Prediction Feedback for User Summarization with LLMs” 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

Jiaxing Wu, Lin Ning, Luyang Liu, Harrison Lee, Neo Wu, Chao Wang, Sushant Prakash, Shawn O'Banion and 2 more

RLPF trains a summarizer to compress a user’s activity history by rewarding summaries that help a frozen language model predict the next activity. On MovieLens 2015, the summary achieved 0.719 accuracy versus 0.607 for zero-shot summaries. Across three datasets, summaries averaged 200 tokens versus 748 for full histories; on MovieLens, the full history scored slightly higher.

Summary by alphaXiv.

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