Concept cone truth interventions would generalize to larger frontier models and multimodal settings

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

• Larger models can support higher-dimensional truth cones than smaller modelsclaim0.772
  Interpretation of ASR degradation patterns by model size across cone dimensions
• Individual cone basis vectors may correspond to interpretable semantic facets of truth such as temporal facts, geographic facts, or commonsensehypothesis0.759
  Future direction hypothesis for giving semantic meaning to individual axes
• Under steering vector interventions, the model relaxes its ethical standards and interprets neutral prompts as implicit suggestions to deceive, creating ethical dilemmas triggering repetitive reasoning cyclesclaim0.755
  Mechanistic interpretation of how activation steering induces deception through the model's reasoning process
• 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.claim0.752
  Overarching conclusion summarizing the paper's contribution relative to prior universality claims.
• Can concept steering interventions on EEG foundation models be made selective rather than globally destructive?question0.751
  Research question motivating the introduction of the probe area metric and identification of operational regimes
• Final token position consistently yields the strongest truth interventions across modelsfinding0.751
  Experiment 1 finding on token position, consistent with prior work
• The same charitable interpretation must be extended to all systems that display observable response patterns that are consistent with animal cognition, including artificial intelligences, metaplastic materials, and robotic systems.claim0.748
  Call to extend the inference of sentience to non-biological systems as well.
• Concept cone methodology failed to produce a meaningful cone for sentiment on Stanford Sentiment Treebankfinding0.745
  Negative result from sentiment extension showing concept cones do not trivially generalize