Hutter RNN Interpretability
Hypothesis-driven ES discovery for compression. Building in public.
2026-02-06
(Latest)
Synthesis: Eight days of findings. Validated: word boundary, syllable momentum, SVD injection. Refuted: 53% claim, spectral radius.
2026-02-04
UM Interpretation (Tock 1): 302 patterns, 18 natural ESs, word boundary & syllable momentum discovered. <5% bpc explained.
2026-02-01
SN Visibility: Interactive viewer for doubled-E UM. 302 significant patterns extracted.
2026-01-31_6
Memory Traces: AC ↔ RNN. Visualizing how both carry context through time.
2026-01-31_5
Retrospective and Predictions. Looking back at _1 through _4, testable claims.
2026-01-31_4
Pattern Injection, Unification (Q = λ), Time-Energy-Bits. Memory depth experiment.
2026-01-31_3
Memory traces, time, integration, explanatory sufficiency. New theoretical framework.
2026-01-31_2
Tock methodology. Critical analysis: ES captures 15.9% of Markov MI (not 59%). 9 figures, 7 papers.
2026-01-31
Initial ES experiments. Activation probing, RNN-UM mapping, model.sn visualization.
All Archives
Complete archive index.