Vision 4-shot cat classification achieves reliable recognition when S(A) exceeds Sc

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

• InceptionV1 neuron 4e:55 responds to cat faces, fronts of cars, and cat legs as unrelated stimulifinding0.727
  Concrete example of polysemantic neuron demonstrating the challenge to the circuits agenda
• Copy-cat processes (partial involutions) are computationally universalfinding0.723
  Mere copying of tokens between paired positions suffices to simulate all partial recursive functions and model higher-order logics.
• Learning cooccurrence statistics in either vision or language domain recovers roughly the same perceptual color representationclaim0.718
  Empirical validation that PMI convergence actually occurs on real data
• Diverse computer vision models trained on visual recognition tasks converge to remarkably similar internal feature representations regardless of architecture, training procedure, or implementation details, closely matching the organization of animal visual cortexfinding0.717
  Empirical evidence for the universality hypothesis cited as supporting the possibility of convergent consciousness-like solutions
• Automated interpretability (Claude 3 Opus) and specificity scoring show SAE features are significantly more interpretable and specific than MLP neurons.finding0.716
  Quantitative comparison supporting SAE utility.
• Layer-wise geometry summaries (Sbmax, AUSN) predict internal few-shot thresholds θ50claim0.716
  Claim that geometry-to-behavior correlates exist
• Calibrated few-shot prompting was a surprisingly weak baseline for truth classification compared to linear probesfinding0.715
  Unexpected finding that behavioral baseline underperforms representational probing approaches
• Asynchronously trained DiffLogic CA shows greater robustness to 10x10 pixel damage than synchronously trained version, measured by sum of absolute differencesfinding0.714
  Quantitative comparison of synchronous vs asynchronous training for noise resilience