Powered by OpenAIRE graph
Found an issue? Give us feedback
ZENODOarrow_drop_down
ZENODO
Preprint . 2026
License: CC BY
Data sources: Datacite
addClaim

SNN-Genesis v10.1: Stochastic Resonance in LLM Reasoning — Cliff Anatomy, Smart Defibrillation, and the Prophylactic Principle

Authors: Funasaki, Hiroto;

SNN-Genesis v10.1: Stochastic Resonance in LLM Reasoning — Cliff Anatomy, Smart Defibrillation, and the Prophylactic Principle

Abstract

Erratum : AI tool attribution was corrected in v12. SNN-Genesis v10.1 extends v10's discovery of stochastic resonance in LLM reasoning with two new experiments that explain WHY and HOW the effect works. Phase 58 (Cliff Anatomy): Dissects the σ=0.15→0.20 phase transition by measuring hidden-state diagnostics. Cosine similarity ≈ 0.50 is identified as the critical threshold — reasoning succeeds when the perturbed hidden-state direction is sufficiently preserved (cos ≥ 0.50) and collapses when noise overwhelms the original direction (cos < 0.46, SNR < 0.5). The transition is gradual, not discontinuous, with a therapeutic window of cos ∈ [0.50, 0.70]. Phase 59 (Smart Defibrillation): Tests conditional noise injection — firing σ=0.15 only after errors (like a cardiac AED). Result: defibrillation HURTS (4% vs 18% baseline, N=50). This establishes the Prophylactic Principle: stochastic resonance is preventive (vaccine), not curative (antibiotic). The noise must be present before reasoning enters a failure basin, not after. From v10 (retained): Orthogonal noise decomposition proving direction×magnitude interaction, classic bell curve (9% → 32% → 0%), large-N replication (N=100, p=8.4×10⁻⁵). 61 pages. All experiments conducted on RTX 5080 with Mistral-7B-Instruct-v0.3 (4-bit quantized). Code: https://github.com/hafufu-stack/snn-genesis

Keywords

Spiking Neural Networks, Tower of Hanoi, Stochastic Resonance, Large Language Models, Liquid Neural Networks, Alignment Tax, AI Safety, Closed-form Continuous-time, Honest Null Result, Chat Template Effect, Deep-Thinking Ratio

  • BIP!
    Impact byBIP!
    selected citations
    These citations are derived from selected sources.
    This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    0
    popularity
    This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
    Average
    influence
    This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    Average
    impulse
    This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
    Average
Powered by OpenAIRE graph
Found an issue? Give us feedback
selected citations
These citations are derived from selected sources.
This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Citations provided by BIP!
popularity
This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
BIP!Popularity provided by BIP!
influence
This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Influence provided by BIP!
impulse
This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
BIP!Impulse provided by BIP!
0
Average
Average
Average
Upload OA version
Are you the author of this publication? Upload your Open Access version to Zenodo!
It’s fast and easy, just two clicks!