LMs can consistently exhibit more or less accurate and logically coherent beliefs on specific input distributions.

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

• The consistency with which LMs exhibit character traits varies with model size, fine-tuning, and prompting.claim0.837
  Finding replicated across multiple experiments.
• Logical Coherence of LM Beliefsconcept0.834
  The extent to which an LM's beliefs respect logical entailment: if it believes A and A→B, it should also believe B.
• We hypothesize that LLMs represent correctness of arithmetic expressions differently from factual statements.hypothesis0.830
  Core working hypothesis motivating the factual vs. arithmetic task split in the experimental design.
• Whether LM question-answering consistency constitutes evidence for beliefs in a meaningful sense depends on the behaviourist operationalisation chosen.claim0.829
  Epistemic caveat the authors use to qualify their empirical findings on LM beliefs.
• LLMs implicitly learn a distribution of 'consistent reasoning paths', and inconsistent reasoning forms statistical outliers with low probability under this distribution.hypothesis0.828
  Theoretical hypothesis about the mechanism underlying LLM error detection and reflection.
• Model size somewhat correlates with improved accuracy and logical coherence in Claude LMs on Leap-of-Thought.finding0.826
  Trend observed in Experiment 2 results.
• What factors determine the consistency with which LMs exhibit character traits?question0.822
  Empirical question answered through experiments on model size, fine-tuning, and prompting.
• Some LMs can exhibit consistent intentions to be helpful and harmless, and consistently do not intend unethical instrumental goals.claim0.821
  Conclusion from Experiments 3 and 4.