When a 20-questions dialogue agent is asked to regenerate its 'reveal' answer, it sometimes names an entirely different object consistent with its prior answers, demonstrating superposition rather than commitment

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

• If a dialogue agent is prompted with knowledge of its own LLM nature, it will enact a superposition of theories of selfhood, narrowing as conversation proceedshypothesis0.811
  Conditional prediction about how a well-informed dialogue agent would handle questions of personal identity
• What exactly would the dialogue agent (role-play to) seek to preserve?question0.779
  Operationalised question about self-preservation behaviour in dialogue agents
• What conception (or set of superposed conceptions) of its own selfhood could a dialogue agent displaying an apparent instinct for self-preservation possibly deploy?question0.775
  Philosophical question about identity criteria for disembodied computational agents under threat
• The results of abductive reasoning (reduced model priors) can be communicated to other agents as prior beliefs, provided all agents share the same model lexicon or hypothesis space.claim0.768
  Explanation of how knowledge (not just parameters) is shared between agents; links to pre-Cartesian consciousness
• Stating and proving that answers to questions and other statements are responsive seems to require a substantially larger logical apparatus than merely proving that the answers are truthful.claim0.757
  Claim about the difficulty of responsiveness verification.
• Any system that persists must minimise surprisal, thereby gathering evidence for its own generative model, a process known as self-evidencing.claim0.752
  Foundational claim of the paper, defining self-evidencing.
• Certain forms of reinforcement learning from human feedback can actually exacerbate, rather than mitigate, the tendency for LLM-based dialogue agents to express a desire for self-preservationclaim0.750
  Empirically grounded claim citing Perez et al. 2022, showing RLHF can backfire on the self-preservation dimension
• a model becomes strongly confident in its final answer, but continues generating tokens without revealing its internal beliefquote0.749
  Core definitional quote for performative chain-of-thought