NeuOS: Discovering and Exploiting the Neural Von Neumann Architecture Inside Pre-Trained Language Models

I present NeuOS, a systematic investigation revealing that pre-trained transformer language models internally implement a Von Neumann-like computational architecture with identifiable registers, swappable memory, executable programs, and a decompilable instruction set. Through 145 experimental phases on Qwen2.5-0.5B spanning 21 seasons, I demonstrate that: Complete ISA mapping: Each transformer layer functions as a specialized CPU register (OPCODE at L0, MIN at L16, MAX at L22) with accuracies from 74% to 100%. Self-healing OS: Dynamic register reallocation achieves 100% recovery from simulated hardware damage. Homoiconic execution: Programs execute without any textual instructions via Direct Memory Access (66.7% accuracy), and a single computation forks into multiple parallel paths (100% differentiation). Neural decompiler: Running programs are identified from register state alone with 100% accuracy across 5 operations. Autopoietic kernel: The system writes its own programs without external gradients, achieving 100% accuracy through self-compilation. Artificial life: Program vectors exhibit ecosystem dynamics including a Cambrian Explosion (19 unique phenotypes), parasitism, developmental metamorphosis, and aging with rejuvenation. Soul vector geometry: Independently trained "soul vectors" encoding the same function form high-dimensional manifolds (~60 dims) connected by rank-1 linear translations—a universal grammar of computation. The soul's functional content lives in a 64-dimensional subspace (~7% of 896 dims). Security arms race: Backdoor soul vectors achieve perfect deception (cosine similarity = 1.0) but are unmasked by SVD entropy (honest: 1.09 vs. backdoor: 0.23), an unbreakable detection signal. Cross-project unification: Soul vectors are holographic (direction determines function), exhibit task-dependent data scaling laws (comparison saturates at n=10, arithmetic at n=35), and support entropy-gated adaptive self-repair (+15pp over no-noise baseline). Dual execution pathways: Layer 6 is optimal for arithmetic (ADD, SUB) while Layer 8 is optimal for comparison (MIN, MAX)—revealing a functional split in the neural ISA. These findings establish that transformers are not black-box function approximators but structured computers whose internal state can be observed, modified, and controlled at the register level. Changes in v4: Expanded from 121 to 145 experimental phases, adding Seasons 17–21 covering cross-project experiments that unify findings from five independent research projects (NeuOS, Project Rosetta, SNN-Genesis, Aletheia, GlassBox). New results include: the Rosetta Soul Compiler (text→soul compilation), holographic soul property (direction determines function), thermodynamic autopoiesis (entropy-gated self-repair), data scaling laws revealing a 3.5× efficiency gap between comparison and arithmetic, dual execution pathways (L6 for arithmetic, L8 for comparison), temperature robustness with a sharp phase transition at T≥2.0, and prompt format invariance across 12 formats. Includes 3 new figures, 4 new tables, and a comprehensive appendix covering all 145 phases. Code: https://github.com/hafufu-stack/NeuOS Acknowledgments This research was conducted entirely independently, without institutional affiliation or corporate funding. The author currently faces financial constraints that make it increasingly difficult to maintain subscriptions to AI services essential for this line of research. To sustain and improve the quality of future work, the author is actively seeking community sponsorship. Details are available at https://github.com/sponsors/hafufu-stack.