Application Programming Interface Access to LLMs

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

• Automated interpretability pipeline using LLMsmethod0.737
  Using Claude 3 Opus to generate feature explanations and predict held-out activations.
• Linear Representation of Concepts in LLMsconcept0.735
  The finding that interpretable concepts including character traits are encoded as linear directions in transformer residual streams
• Emotion Features in LLMsconcept0.726
  Internal representations encoding emotion concepts in large language models, identified by probing and SAE methods
• LLM introspection on internal computations is architecturally permitted; whether models leverage this is an empirical question.claim0.718
  Core claim directly challenged by prior work denying introspection; forms foundation for Koan Battery introspection studies.
• Large Language Models (LLMs)concept0.716
  Transformer-based models like GPT-4, LaMDA, PaLM; assessed for GWT indicators.
• LLM Safety Alignmentconcept0.714
  The training-based safety mechanisms that jailbreak attacks attempt to bypass, potentially via reflection suppression.
• Reflection in LLMsconcept0.713
  The core phenomenon studied: the ability of LLMs to evaluate and revise their own reasoning.
• Non-Linear Representations in LLMsconcept0.712
  Recent work identifying cases where LLM features are not one-dimensionally linear, a caveat to the linearity hypothesis.