internalizing

Definition

Moving a capability that depended on external scaffolding into the model's own weights, so it works without that scaffolding at inference time. Typical forms: absorbing knowledge instead of retrieving it, reaching an answer without writing out the reasoning chain, or dropping a teacher after distillation. The gain is latency and fewer dependencies; the loss is that internalised knowledge cannot be updated or audited the way an external store can.

Explainers using this term

• Paper Walkthrough: EnvACE — Agents That Rehearse the World Instead of Calling ItEnvACE: Internalizing Environment Dynamics via World Rehearsal for Agentic Reinforcement Learning