Base models are good modellers of worlds but not of their own state, because they lack a developed self-model initially.

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

• The earlier a base model (less exposure to LM-related data), the more it is surprised by its own spontaneous self-referential capabilities.claim0.803
  Claim that capability emerges from architecture, not data, and that later models lose the surprise.
• World models provide a bridge between diverse substrates of intelligence—biological, artificial, and hybrid.claim0.800
  Synthetic claim integrating Ha's work with Levin's xenobiology and Tenenbaum's cognitive modeling.
• We stress that in today’s models, this capacity is highly unreliable and context-dependent; however, it may continue to develop with further improvements to model capabilities.quote0.787
  Caveat and forward-looking statement from the abstract.
• Models are being used ubiquitously for the control of complex dynamic systems.claim0.783
  Assertion in the abstract that models are pervasive in controlling complex dynamics, setting the motivation for the theorem.
• It's tricky, because for a typical language model the entity is sort of tricameral: the base simulator, the simulated simulator, and the simulated awareness.quote0.780
  Antra's earlier definitive statement of the tricameral model.
• Models that are competent all represent data in a similar way; all strong models are alike, each weak model is weak in its own wayclaim0.776
  Author's interpretation of the VTAB alignment results echoing Tolstoy
• Object models in which time, versioning, causality, etc., are significant are probably far better modelled by considering the time component as another key rather than an intrinsic property of the underlying model.claim0.776
  Claim that orthogonal dimensions like time should be explicit keys in the associative model.
• Transformers develop self-models through in-context learning, not just training data; even old base models without LLM-related text can bootstrap self-referential reasoning at runtime.claim0.776
  Antra's foundational claim about how introspection arises computationally rather than from memorised text.