A formal behaviourist framework can describe LM behaviour precisely in intuitive language without undue anthropomorphism.

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

• How should humans talk about language models without undue anthropomorphism?question0.779
  Motivating question for the whole paper.
• Understanding how LMs learn linguistic behaviours may offer insights into fundamental properties of languagehypothesis0.778
  Forward-looking hypothesis linking LM mechanism analysis to linguistic theory
• If the conceptual framework we use to understand other humans is ill-suited to LLM-based dialogue agents, what alternative conceptual framework should we use?question0.769
  The motivating question the paper sets out to answer by proposing role play and simulation metaphors
• Anthropomorphism of LMsconcept0.765
  The problematic practice of attributing human-like beliefs, intentions, and emotions to language models without formal grounding.
• LLMs require hierarchical or embodied substrates; stigmergy via environment interaction is the biological solution.claim0.765
• Whether LM question-answering consistency constitutes evidence for beliefs in a meaningful sense depends on the behaviourist operationalisation chosen.claim0.764
  Epistemic caveat the authors use to qualify their empirical findings on LM beliefs.
• Some LMs can exhibit consistent intentions to be helpful and harmless, and consistently do not intend unethical instrumental goals.claim0.759
  Conclusion from Experiments 3 and 4.
• LLMs trained only on language data have rich enough knowledge of visual structures that decent visual representations can be trained on images generated solely by querying the LLMfinding0.757
  Sharma et al. result supporting cross-modal alignment: language-only models implicitly encode visual structure