Reproduce “In-Parameter Learning: Why Lifelong AI Systems Need More Than Longer Context” 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
Yiding Wang, Haotong Yang, Pingzhi Tang, Shijia Kang, Cai Zhou, Chenglin Zhu, Fan Jiang, Fanxu Meng and 15 more
The authors argue lifelong agents need IPL alongside ICL. ICL uses inference-time information without parameter changes; IPL consolidates selected experience into parameters during deployment. They propose context for immediate information and parameters for durable learning. The paper estimates lifetime spoken conversation at 4 × 10⁸ tokens, two orders beyond a 1-million-token window; this position paper does not test a system.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.