Skip to slide
34 / 53

CH 05 · Recursive Language Models, Thinking Beyond the Memory Limit · 01/07

Read as one page

Recursive Language Models, Thinking Beyond the Memory Limit

Recursive language models keep a huge input outside the model as explorable data, then write code to inspect pieces and call fresh copies on the rest.

7 slides · 4 min read

Contour map for Recursive Language Models, Thinking Beyond the Memory Limit

In this chapter

  1. 01The problem: long inputs break models
  2. 02The idea: treat the long prompt as an environment, not a thing to read
  3. 03The recursive part
  4. 04Why this works so well
  5. 05Why this paper mattered
  6. 06The one-sentence takeaway

Paper: Recursive Language Models (2025)

Every chapter so far made the model think better. This final chapter tackles a different wall: thinking about something bigger than the model can hold. Models have a fixed context window, a maximum amount of text they can read at once. Feed them a whole book, a giant codebase, or a thousand-page report, and they either cannot fit it or, just as bad, get confused and lose accuracy as the input grows. Recursive Language Models, or RLMs, are a clever way around this, and they come from MIT researchers in 2025.

Finished this idea?

Your learning progress stays in this browser.