Pass-Through Gate Bias

Initial gate distribution biased toward pass-through gates A and B to facilitate training stability

Neighborhood — ranked by edge-count

concept

Probabilistic Gate Selection
associated_with
Each gate maintains a 16-dimensional probability distribution over binary operations, updated via gradient descent

cosine ≥ 0.65 · no typed edge

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

Bias Amplificationconcept0.709
Problem cited as a limitation of current LLMs; PRH predicts larger models should amplify bias less
Different tasks yield distinct distributions of logic gate types between perception kernels and update networksclaim0.709
Interpretive claim based on circuit analysis across experiments
Pre-Encoder Biasconcept0.708
Architectural modification subtracting a learned bias from autoencoder inputs before encoding; initialized to geometric median of dataset; improves autoencoder performance
gatesconcept0.704
Bias in language modelsconcept0.694
Features related to gender, racial, ethnic biases, slurs, and hate speech.
Logic Gate Perception Kernelsmethod0.692
Named technique in DiffLogic CA where fixed-structure logic circuits replace Sobel filters for neighborhood perception
Learned Google G circuit uses 927 active gates (excluding pass-through gates A and B)finding0.691
Circuit complexity metric for colored pattern experiment, largest of all experiments
Simplicity Biasconcept0.688
The tendency of deep networks to implicitly favor simpler solutions that fit the data, driving convergence