question
active
question:how-can-the-stability-and-variance-of-feature-discovery-across-training-runs-be-improvedHow can the stability and variance of feature discovery across training runs be improved?
Future work.
Source paper
extracted_from(2026) · Thomas Jiralerspong · Trenton Bricken
Neighborhood — ranked by edge-count
Papers (1)
paper
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.
- How do features change over model training; do they form abruptly or gradually, and do their directions drift?question0.779Fresh-traction question.
- Forward-looking interpretation of scaling results.
- Scaling laws for dictionary learning are unknown and needed to assess feasibility on frontier models
- Empirical principle discovered during autoencoder training; led to using 8 billion training points
- Optimal number of features scales faster than optimal number of training steps with compute budget.finding0.768Allocation result from scaling laws.
- Ethical implication about the nature of AI training experience if the thesis holds
- Future work question arising from the finding that Combined model did not outperform MNLI model
- Mechanism from Theorem 4.