English

When Self-Reference Fails to Close: Matrix-Level Dynamics in Large Language Models

Computation and Language 2026-04-15 v1

Abstract

We investigate how self-referential inputs alter the internal matrix dynamics of large language models. Measuring 106 scalar metrics across up to 7 analysis passes on four models from three architecture families -- Qwen3-VL-8B, Llama-3.2-11B, Llama-3.3-70B, and Gemma-2-9B -- over 300 prompts in a 14-level hierarchy at three temperatures (T{0.0,0.3,0.7}T \in \{0.0, 0.3, 0.7\}), we find that self-reference alone is not destabilizing: grounded self-referential statements and meta-cognitive prompts are markedly more stable than paradoxical self-reference on key collapse-related metrics, and on several such metrics can be as stable as factual controls. Instability concentrates in prompts inducing non-closing truth recursion (NCTR) -- truth-value computations with no finite-depth resolution. NCTR prompts produce anomalously elevated attention effective rank -- indicating attention reorganization with global dispersion rather than simple concentration collapse -- and key metrics reach Cohen's d=3.14d = 3.14 (attention effective rank) to 3.523.52 (variance kurtosis) vs. stable self-reference in the 70B model; 281/397 metric-model combinations differentiate NCTR from stable self-reference after FDR correction (q<0.05q < 0.05), 198 with d>0.8|d| > 0.8. Per-layer SVD confirms disruption at every sampled layer (d>+1.0d > +1.0 in all three models analyzed), ruling out aggregation artifacts. A classifier achieves AUC 0.810.81-0.900.90; 30 minimal pairs yield 42/387 significant combinations; 43/106 metrics replicate across all four models. We connect these observations to three classical matrix-semigroup problems and propose, as a conjecture, that NCTR forces finite-depth transformers toward dynamical regimes where these problems concentrate. NCTR prompts also produce elevated contradictory output (+34+34-5656 percentage points vs. controls), suggesting practical relevance for understanding self-referential failure modes.

Keywords

Cite

@article{arxiv.2604.12128,
  title  = {When Self-Reference Fails to Close: Matrix-Level Dynamics in Large Language Models},
  author = {Ji Ho Bae},
  journal= {arXiv preprint arXiv:2604.12128},
  year   = {2026}
}

Comments

14 pages, 4 figures, 11 tables

R2 v1 2026-07-01T12:07:42.911Z