method
active
method:stepwise-masStepwise MAS
MAS variant applying interchange interventions at multiple contiguous token positions from the start of a sequence to a sampled time step t.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Model Alignment Searchintroduces
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.
- Novel method that applies intervention only when the model begins a new thinking step (at the \n\n delimiter) rather than at every token
- Design method: take small steps, deciding only what is known with certainty; reject guesses and large-scale trial-and-error.
- The primary contribution of the paper: a bidirectional causal method that learns rotation matrices for each model to uncover and compare causally relevant latent subspaces across neural networks.
- The idea that living structure emerges only through a sequence of small, structure-preserving moves, not by a single grand blueprint.
- Design approach emphasizing breaking wholes into parts while preserving relationships, praised by a designer as Alexander's 'gold'.
- Segments of reasoning separated by \n\n tokens used as the unit of analysis in ReflCtrl
- A 12-step sequence of structure-preserving steps proposed in this chapter for shaping urban space as positive hulls, beginning with identifying main spaces and ending with subdividing interiors.
- The incremental unfolding characteristic of morphogenesis, where each step arises from the previous state.