method
active
method:local-replacement-modelLocal replacement model
CLT plus error nodes and original attention patterns, local to a prompt.
Neighborhood — ranked by edge-count
Papers (2)
paper
Frameworks (1)
framework
- MLP replacement with sparse features underlying the original attribution graphs.
Concepts (1)
concept
- Error nodesusesUninterpretable nodes representing replacement model reconstruction error.
Methods (1)
method
- Replacement model architecture with 30M features replacing MLP neurons.
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.
- Substitutes all MLP outputs with CLT reconstructions.
- When a single conversation involves different models, e.g., routing between standard and reasoning models
- A representation that captures relevant aspects of a system; according to the theorem, the regulator must embody this.
- Primary test domain for manifold steering, including reasoning and ICL tasks
- Later layers correct single-layer steering interventions.
- Large pretrained models used as backbones across tasks; their universality motivates the convergence hypothesis
- Comparing models using log-evidence approximated by free energy.
- A model trained on comparison data to assign scores to responses, used as reward signal in RLHF/RLAIF.