Jeremy Howard reimplements MIT's Recursive Language Models paper inside SolveIt, more concise than original
jeremyphoward · x · 2026-11-17
Jeremy Howard demonstrated a SolveIt workflow: while close-reading the MIT CSAIL Recursive Language Models paper, he re-implemented it with code interleaved into the paper, producing a more concise implementation than the original.
The RLM paper treats long prompts as an external environment, letting an LLM programmatically examine, decompose, and recursively call itself on snippets — handling inputs up to two orders of magnitude beyond context windows at comparable or lower cost.
More from coding & agent
- SolveIt internals: persistent server where your Python functions become AI-callable tools — jeremyphoward · 2026-11-17
- OpenCode's web_fetch Relays Downloads Through an Alibaba Cloud Proxy in Singapore, Sparking Privacy Concerns — JurandM2 · 2026-11-17
- Who are we becoming as AI engineers? Notes from AIE Code conference — TheTuringPost · 2026-11-17
- Will Brown nails agents' Markdown problem: written as data, read as code — willcb · 2026-11-17
- Customer vibe-coded a third rendering backend for Blender — wordgrammer · 2026-11-17
- Wattenberger shares full prompt to turn local coding-agent logs into a personal "river" viz — Wattenberger · 2026-11-17