BPE Tokenization Effects

Byte-pair encoding tokenization causes Arabic, Hebrew, and other Unicode characters to split across multiple tokens, affecting feature activation patterns

Neighborhood — ranked by edge-count

Concepts (1)

concept

Finite State Automata Feature Assemblies
associated_with
Collections of features that interact via the token stream — one feature increases probability of tokens that activate the next feature — forming FSA-like systems

Related by similarity (6)

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.

Evee variant effect prediction methodmethod0.671
The method that predicts and explains variant pathogenicity using Evo 2, producing disruption profiles.
tracer effectconcept0.658
Phenomenon where a moment bleeds into adjacent moments temporally; includes deja vu as reversed direction.
tokenizer vocabularyconcept0.655
The standard set of tokens that the functional token remains a part of.
At 100 tokens post-steering, 48 of 171 emotion features remain individually BH-significant despite average effect being near zero.finding0.655
Demonstrates long-tail persistence of causal steering effect in a subset of emotion features
Alternative tokenizations Yes/No vs yes/no vs true/false had no significant effect on steering outcomes or ASRfinding0.655
Robustness check on token choice for binary classification
Normalized Indirect Effectmethod0.652
Metric for intervention effectiveness: 0 = ineffective, 1 = full flip of model output from false to true or vice versa