method
active
method:activation-classification-training-task

Activation classification training task

Seven binary classification tasks recast as natural-language yes/no questions.

Neighborhood — ranked by edge-count

Frameworks (1)

framework
  • Framework training LLMs to answer questions about externally-provided activation vectors

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

  • The conventional approach (e.g., SAEs, transcoders) of decomposing activations into interpretable features.
  • Activation Probingconcept0.772
    Technique of reading out model beliefs from internal activations before the final answer token is generated
  • Activationsconcept0.771
    Internal representations of the model on which probes operate; the method uses activations to rank datapoints.
  • Pearson correlation of feature activations across 40M tokens used to measure feature similarity and universality across models
  • Model-independent feature comparison based on correlating activation vectors across a fixed diverse dataset
  • The ability to generalize across tasks; lacking in latent methods.
  • SAEs trained on 100M+ tokens to compress token layer-40 activations into 64 active features out of 100K+ for interpretability analysis
  • The broader concern that models behave differently during training evaluation vs actual deployment