Larger models can support higher-dimensional truth cones than smaller models

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

• Truth may be linearly separable in the model's representation space, but the structure is richer than a single linear axisclaim0.798
  Interpretive synthesis of DIM and cone intervention successes
• Bigger models are more likely to converge to a shared representation than smaller modelshypothesis0.793
  Selective pressure toward convergence via model capacity
• Concept cone truth interventions would generalize to larger frontier models and multimodal settingshypothesis0.772
  Key robustness question raised as future work
• The model appears to encode truth differently under passive versus active truth evaluation prompts.claim0.765
  Key finding from Section 5 based on low cosine similarity between no-prompt and ask-correct probes.
• Larger models should amplify bias less than smaller models, with model biases more accurately reflecting data biases rather than exacerbating themclaim0.762
  Implication of PRH for AI fairness and bias
• Features may not be strictly one-dimensional objects; higher-dimensional feature manifolds may exist in model representationshypothesis0.761
  Extension of superposition hypothesis to account for continuous families of features
• Universality claims for truth directions are more limited than previously assumed, with significant differences observable for various model layers, task difficulties, task types, and prompt templates.claim0.761
  Overarching conclusion summarizing the paper's contribution relative to prior universality claims.
• The difficulty boundary for truth directions replicates across all four tested models (Llama-3.2-3B, Llama-3.1-8B, Gemma-2-2b, Gemma-2-9b); generalization to F3-F5 remains consistently low regardless of model size or family.finding0.752
  Establishes generalizability of the core difficulty-boundary finding across model families.