concept
active
concept:anti-correlated-headsAnti-correlated Heads
Attention heads whose output direction is opposite to the aggregate attention persona vector; hypothesized to maintain coherency
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.
- Motivates including Head Cor+Anti condition in steering position comparison
- Hypothesized heads that invert semantic meaning of features; not confirmed.
- QK circuit heads hypothesized to measure likelihood of an output given prior activations, used in prefill detection.
- Transformer attention heads that could be recruited to extract different kinds of information (text vs. thoughts).
- Mechanistic circuits in transformers documented by Olsson et al. 2022, cited as evidence for pattern-repository assumption
- Possibility that functional units are distributed across heads.
- m=2 experiments.
- Behavior of IOI inhibition head and GPT-2 L10H7.