Fine-tuned LMs displayed higher mean HH-intent scores and increased consistency compared to pre-trained counterparts.

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

• LMs can consistently exhibit more or less accurate and logically coherent beliefs on specific input distributions.claim0.815
  Conclusion from Experiment 2 on Leap-of-Thought.
• Larger LLMs show greater reduction in deceptive behavior after SOO fine-tuningfinding0.799
  Scaling pattern: 78B > 27B > 7B in deception reduction from SOO fine-tuning
• The consistency with which LMs exhibit character traits varies with model size, fine-tuning, and prompting.claim0.798
  Finding replicated across multiple experiments.
• Smallest models have the lowest HH-intent scores, in accordance with their relative weakness at reasoning and adaptation.finding0.794
  Main result from Experiment 3 on HH-intent scaling with model size.
• SOO fine-tuning significantly reduces deceptive behavior in LLMs while maintaining general task performanceclaim0.791
  Central empirical claim of the paper supported by three LLM experiments
• Hypothesis: Fine-tuning reduces mismatch dr between prior and targethypothesis0.786
  UCCT's theoretical prediction about how fine-tuning maps onto the anchoring score
• Models fine-tuned on human data show strong positive correlation between misalignment and incoherence; misalignment scores fall below 1% when incoherence threshold is appliedfinding0.785
  Human data fine-tuning effect is distinct from synthetic emergent misalignment and likely caused by off-policy training
• SOO fine-tuning could complement RLHF and Constitutional AI by fostering internal coherence that promotes honest behaviorsclaim0.784
  Integration claim positioning SOO as additive to existing alignment approaches