hypothesis
active
hypothesis:useful-non-robust-features-could-be-an-artifact-of-superpositionUseful non-robust features could be an artifact of superposition
Open question linking to Goh.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
Thinkers (1)
thinker
- Gabriel GohstudiesReported multimodal neurons in vision models including region neurons and person detector neurons
Questions (1)
question
- Open question.
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.
- Posed on SAE results.
- Core mechanism statement.
- Discussion implication.
- features serve as oversight signals to intentionally design models with desirable capabilitiesquote0.760Core vision statement.
- Robustness influences superposition, predicting lower SAE loss in more robust models.hypothesis0.757Popperian credit for surprising prediction.
- Strategic picture.
- Open question in limitations.
- Mechanistic finding by Bricken et al. 2023 about how LLMs store features; cited as operational justification for pattern-repository assumption