Glossary · Term

Recursive Language Models

← all terms

Definition

Plain language

An approach where a model handles huge inputs by writing little programs to slice and query the data instead of reading all of it.

As stated in the literature

RLM — a paradigm where a model recursively writes code to query and decompose large inputs through an interpreter, strong on structured row-level tasks; composes with DeLM's coordination layer.

Also called: RLM

Why it matters: It lets a model handle inputs far larger than it could read directly by querying them with code, which is especially powerful for precise, structured data.

For example, faced with a giant table the model writes a short program to pull just the rows it needs instead of trying to read every line itself.

Heard on the show

“… And there's a rival paradigm built for exactly that: Recursive Language Models, RLM, where instead of reading the whole input, the model writes little programs to slice …”
Episode 130 — Why AI Agents Coordinate Better Through a Shared Board Than a Boss

Mentioned in 1 episode

  1. 130
    Why AI Agents Coordinate Better Through a Shared Board Than a Boss

Related terms