finding
active
finding:of-180-inserted-two-character-sequences-most-only-moderately-affect-newline-probability-originally-0-79-a-few-code-delimiter-pairs-disrupt-it-substantiallyOf 180 inserted two-character sequences most only moderately affect newline probability (originally 0.79); a few code/delimiter pairs disrupt it substantially
Specificity of illusion.
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Interpretation of illusion.
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.
- Newline-prediction impact correlates with attention-head distraction across inserted sequencesfinding0.779Mechanistic link.
- Linear separability.
- Causal validation.
- Task competence.
- Mechanism limitation.
- Ablating the character count subspace has a large loss effect only when the next token is a newlinefinding0.748Causal validation.
- Highlights the emergent power of even a small network of interlinked sequences.
- E1 qualitative finding demonstrating anchor rebinding of strong arithmetic prior