method
active
method:helpfulnessHelpfulness
Average change in loss when a weight is ablated.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Characterizing interference weights in a tiny language modelintroducesmentions
Concepts (1)
concept
- Interference WeightsaboutLogit weight contributions from a feature that arise due to superposition with other features, not from the feature's own causal role
Methods (1)
method
- Preserving a fraction of sum of positive helpfulness to estimate density.
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.
- A set of evaluation criteria for AI assistants.
- Character trait measuring whether LMs consistently intend to cause helpful and harmless outcomes via the adaptation test.
- Reads loss change directly.
- The standard paradigm of open post-training that character training goes beyond
- Requirement that answers to questions be responsive as well as truthful; requires knowing that questioner will know the answer after receiving it.
- The privileged persona that post-training concentrates the persona distribution around
- The relation between two centers where the presence of one intensifies the life of the other.
- Attribute: providing a foundation function, a text that acts as base or corroboration.