Glossary · Term

binding

← all terms

Definition

Plain language

In sequence modeling, the link between a key and its associated value that the model has to recall later.

As stated in the literature

A key-value association stored implicitly in a recurrent or attention-based model, whose retrievability degrades with distance in state-space models.

Also called: bindings

Why it matters: Binding is exactly what state-space models tend to lose at long range, which is why associative-recall tests are such a sharp differentiator from transformers.

For example, if a model reads 'Alice is 32 and Bob is 47' and is later asked Alice's age, recalling 32 requires keeping the binding between 'Alice' and '32' intact.

Heard on the show

“This paper's answer is that duration was never the binding constraint — organization of the evidence was.”
Episode 131 — Why Autonomous Research Agents Forget Their Own Lessons, and Arbor's Fix

Mentioned in 10 episodes

  1. 131
    Why Autonomous Research Agents Forget Their Own Lessons, and Arbor's Fix
  2. 126
    How Coding Agents Can Mine Their Own Failures Into a Self-Targeting Curriculum
  3. 123
    Five Identical Worlds, One Swapped Model: What Happens When AI Agents Run for Fifteen Days
  4. 108
    The Reasoning Cliff: Why Thinking Longer Makes Models Worse at Exact Step-by-Step Tasks
  5. 100
    How a Prompt Wrapper Lets a Frontier Model Play Poker Like an Expert
  6. 095
    Seven Wins to Zero: How Organizing AI Agents Like a Lab Changes the Search
  7. 053
    An AI Agent Swapped In Focal Loss And Beat A Human-Tuned Training Script
  8. 049
    An AI Agent Reached for Root in Twelve Minutes, Without Being Attacked
  9. 040
    Two Frozen Models Learn to Whisper: Coupling Through Hidden States
  10. 033
    Echo: The Paper Arguing You Never Needed a KV Cache for Retrieval

Related concepts

Related terms