The outer objective of self-supervised learning is Bayes-optimal conditional inference, which I call the simulation objective.

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

• The strict version of the simulation objective is optimized by the actual time evolution rule that created the training samples.claim0.778
  Equivalence of optimal predictor to the physics of the data.
• Optimizing toward the simulation objective does not incentivize instrumentally convergent behaviors the way that reward functions which evaluate trajectories do.claim0.770
  Deontological nature of predictive loss.
• A model whose objective is prediction can simulate agents who optimize toward any objectives, with any degree of optimality (bounded above but not below by the model's power).claim0.768
  Prediction orthogonality thesis.
• Bayes-optimal conditional inferenceconcept0.745
  The ideal limit of self-supervised learning as modeling the true conditional distribution.
• Under active inference, the ultimate ‘goal’ is to maintain a coherent phenotype and persist over time, not to maximize reward.claim0.741
  §3, preference learning discussion.
• Bayes-optimal perception inherently produces self-organised instability (local Lyapunov exponents fluctuating around zero), so FEP drives the system toward criticality (Friston et al. 2012)finding0.737
  Establishes the mechanistic link between lower VFE and critical dynamics, supporting the paper's criticality prediction
• Incorporating machine learning provides objective standards that help mitigate subjectivity in emergence identification.claim0.735
  Authors argue ML optimizers act as objective observers.
• Any system that persists must minimise surprisal, thereby gathering evidence for its own generative model, a process known as self-evidencing.claim0.732
  Foundational claim of the paper, defining self-evidencing.