Response-Average Token Extraction

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

• Response-average token extraction yields stronger steering effectiveness than prompt-last or prompt-average extraction positionsfinding0.817
  Justifies the choice of response tokens for persona vector extraction in the pipeline
• One-Token Likert Rating Extraction Protocolmethod0.774
  Protocol decoding one token and accepting if valid Likert rating, retrying up to 10 times before generating additional tokens
• In A/4, over 100 features primarily respond to the token 'the' in different contextsfinding0.718
  Demonstrates prevalence of token-in-context features and feature splitting of common tokens
• Tokenconcept0.712
  Basic unit of LLM input/output: words, parts of words, punctuation marks, emojis
• Value extractionconcept0.708
  The initial stage of uncertainty metabolization, pulling usable value from sensations.
• PCA Analysis of Token Embeddings/Unembeddingsmethod0.705
  PCA applied to token embedding and unembedding matrices to understand what fraction of residual stream dimensions they occupy and how they relate
• Summarization Token Behaviorconcept0.702
  Behavior where information about full clauses is encoded over clause-ending punctuation tokens in LLMs
• Response Text Augmentationmethod0.699
  Strategy using GPT-4o, Claude 3.5 Sonnet, and Gemini to generate additional responses preserving original meaning, targeting ≥1000 words concatenated per score category.