Pure binary logic circuit inference is exceptionally fast on any hardware compared to continuous neural network inference

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

• Synchronously trained DiffLogic CA circuit succeeds at asynchronous inference without retrainingfinding0.764
  Unexpected result demonstrating robustness of learned circuits beyond their training regime
• Different tasks yield distinct distributions of logic gate types between perception kernels and update networksclaim0.739
  Interpretive claim based on circuit analysis across experiments
• Active inference offers a degree of plausibility as a process theory of brain function.claim0.724
  Overall assessment from Discussion.
• Circuits could act as an epistemic foundation for interpretability by breaking down model behavior into falsifiable statements about small subgraphs.claim0.724
  Normative vision for how the circuits agenda could resolve the pre-paradigmatic state of interpretability
• Neural networks compute cyclic concepts in generic substrate machinery (base-10 addition) not naturally cyclic computation.claim0.722
• Sparse low-cardinality circuits implement competence; 0.2% of neurons handle shared computation across all cyclic tasks.claim0.717
• Chain-of-thought reasoning improves large model accuracy on HHH binary comparisons, reaching ~78% for 52B model, competitive with human-feedback PM.finding0.710
  Figure 4 shows CoT improves over zero-shot, and ensembled CoT further boosts accuracy.
• Neural bioelectric networks achieve a critical separation of hardware and software: the same molecular substrate can instantiate different informational content based on history, enabling minds to arise from matter.claim0.709
  Foundational for understanding how physiology becomes meaning; decoupling of material state from information content is prerequisite for emergence of cognitive Self.