Recursive Language Models, presented by Jordan Henkel
Details
We're back!
Gen AI is on everyone's mind, so what better way to spend a Tuesday evening than learning about the latest techniques from a deep expert?
Paper Summary Summary:
Recursive Language Models (RLMs) let LLMs handle prompts far beyond their context window by treating long inputs as an external environment and recursively reading, decomposing, and re-calling themselves on chunks. Across long-context tasks, RLMs outperform standard frontier models and common scaffolds at similar cost, with RLM-Qwen3-8B delivering large gains over the base Qwen3-8B and nearing GPT-5 on several benchmarks.
About the speaker: Jordan Henkel is an AI/ML Researcher at Sema4.ai, where he works on building enterprise AI agents and translating cutting-edge language model research into production-ready systems. He earned his Ph.D. in Computer Science from the University of Wisconsin–Madison, advised by Prof. Tom Reps, where his research focused on the intersection of software engineering, machine learning, and programming languages. Jordan was named a 2021 Microsoft Research Fellow — one of just ten fellowships awarded that year — and previously served as a Senior Scientist at Microsoft's Gray Systems Lab, where he applied ML to systems programming and software engineering. His published research spans program understanding, program analysis, and the robustness of code models.