Definition
Plain language
Storing what an AI learns by changing its own internal weights, rather than by saving text it can search later.
As stated in the literature
A memory channel that writes experience into a model's parameters (e.g., a trainable adapter) at test time, contrasted with prompt-space memory like summaries or retrieval that leaves the weights frozen.
Why it matters: It lets an agent absorb experience directly rather than re-reading saved notes, but mistakes can become baked in rather than easily deleted.
For example, instead of jotting a user's preference into a notes file, the agent slightly adjusts its own internal weights so it simply 'knows' it going forward.
Heard on the show
“… reliable in exactly the regime where you don't need it — when the answer is already inside its parametric memory and search is just a confidence ritual. …”Episode 092 — When Search Agents Don't Really Search: The Memory Shortcut Hiding in Browsing Benchmarks