claim
active
claim:memorization-can-be-a-mechanism-supporting-abstraction-retaining-instances-longer-lets-models-accumulate-signal-to-learn-generalizable-structureMemorization can be a mechanism supporting abstraction: retaining instances longer lets models accumulate signal to learn generalizable structure
Discussion and App. A.
Source paper
extracted_from(2026) · Jing Huang · Daniel Wurgaft · Rachit Bansal · Laura Ruis +5
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Fig. 6b.
Questions (1)
question
- Future work for frequent tasks.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Are there model structures that account for memorization and generalization across large swaths of training data?question0.820Aggregate analysis motivation.
- Selective pressure toward convergence via task generality
- Opus 4.1 demonstrates highest introspective awareness on abstract nouns (justice, peace, betrayal) with nonzero awareness across all concept categories tested.
- Hypothesis box in §3.2.
- Broader interpretive claim about LM learning bias inferred from the findings
- Opening sentence defining self-evidencing.
- Methodological claim about the scientific value of combining causal abstraction with representational geometry analysis
- Memory transfer across tissues and through metamorphosis supports persistence of Self.