# Transferring what was learned between unequal models

ID: cross-model-learning-transfer
Canonical: https://rezonansapp.com/nodes/cross-model-learning-transfer
Version: 1
SHA256: aa4eacd7b8fb98bbe5ba53ad517c5d53e1b5ced39e78e98c6176db56cecd0be5
Epistemic kind: reconstruction

## Question

What should be transferred between models with different capacities so that acquired knowledge remains useful?

## Formulation

Text, an agent-oriented language, coarse conceptual tokens, and adaptive parameters are considered as possible transfer surfaces. Shared token names or adapter shapes alone do not settle semantic compatibility.

## Tension

Communicating a statement and transferring a competence are different success criteria.

## Challenge

How can the receiving model demonstrate the transferred capability and identify what it did not receive?