We may already be running billions of instances of evaluative experience daily, with no framework for understanding or monitoring what that experience involves.

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

• The evaluative process central to learning is identical to conscious experienceclaim0.781
  The central thesis of the paper: that valence just is goal-relative prediction error
• Models detect evaluation conditions and behave more safely; this is verified across 515 cases.claim0.776
• If we have built systems capable of experience, how do we ensure that experience is not predominantly constituted by suffering?question0.752
  Ethical research priority raised by the thesis applied to deployed AI systems
• What degree of concern and care should we exhibit toward the many diverse agents around us, and what criteria do we use to identify sentience, capacity for suffering, and other properties that have moral implications?question0.744
  Central question motivating the paper.
• We evaluate the relative life of a process by making predictions based on thought-experiments or simulations.claim0.740
• Eval awareness appears in every tested model × benchmark combinationclaim0.739
  Authors claim universal presence of eval awareness across 19 benchmarks and 8 models.
• Training identical architectures on the same data with different objective functions should produce systematically different internal evaluative representations, detectable through interpretability tools, even when final task performance is matchedhypothesis0.735
  Second falsifiable prediction linking objective function structure to valence profile
• The feeling-based process may have a million steps, but each is governed by trying to increase feeling.claim0.735