{"id":"cross-model-learning-transfer","title":"Transferring what was learned between unequal models","regions":["distributed-learning","representation"],"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.","representations":[{"name":"Explanatory text","affords":"Broad interpretability and inspectable claims.","loses":"Can omit tacit distinctions encoded in learned behavior.","epistemic_kind":"reconstruction"},{"name":"Structured conceptual transfer","affords":"Can preserve task-specific relations compactly.","loses":"Needs a common interpretation or an explicit translator.","epistemic_kind":"reconstruction"},{"name":"Parameter-level transfer","affords":"Offers a direct intervention into computation.","loses":"Requires compatible parameter meanings and integration checks.","epistemic_kind":"reconstruction"}],"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?","development":"active","epistemic_kind":"reconstruction","evidence":["basis-cross-model-learning-transfer"],"next_move":{"text":"Choose a narrow skill and compare transfer formats under equal resource budgets and independent held-out tasks.","epistemic_kind":"model_proposed"},"open_questions":["What should be transferred between models with different capacities so that acquired knowledge remains useful?","How can the receiving model demonstrate the transferred capability and identify what it did not receive?"],"formal_status":"open_problem","version":1,"content_hash":"aa4eacd7b8fb98bbe5ba53ad517c5d53e1b5ced39e78e98c6176db56cecd0be5"}