large language models

From U(1) to U(5): A Bottom-Up Reconstruction of Gauge Unification

We present a unified bottom-up construction of gauge symmetries from U(1) to U(5), demonstrating that each group U(n) ≅ [U(1) × SU(n)]/Z_n generates an additional gauge boson whose physical identity is fixed by anomaly cancellation alone. Starting from U(3), the extra U(1) is uniquely identified as B−L with quark charg...

From Event Logs to Process Performance Indicators via Declarative Process Mining

In-Context Virtual Machine: Thermodynamic Entropy Bounds and Deterministic State Orchestration in Transformer-Based Autonomous Agents

Autonomous agents built upon large language models (LLMs) frequently suffer from state drift, cumulative hallucination, and non-deterministic divergence across multi-turn reasoning and execution horizons. In this work, we formalize this failure mode from a statistical physics and information-theoretic perspective as th...

In-Context Virtual Machine: Thermodynamic Entropy Bounds and Deterministic State Orchestration in Transformer-Based Autonomous Agents

Autonomous agents built upon large language models (LLMs) frequently suffer from state drift, cumulative hallucination, and non-deterministic divergence across multi-turn reasoning and execution horizons. In this work, we formalize this failure mode from a statistical physics and information-theoretic perspective as th...

Large Language Models for Research Quality Prediction: Theoretical Issues

DataNomics™ Vault App: Sovereign Multi-Agent Gated Repository, Automated Ingestion Countermeasures, and Cross-Thread Cognitive Synchronization Protocols [GenSI-Ref: DN-VAULT-2026-0920]

ABSTRACT, GLOBAL PRIOR-ART AXIS, & SOVEREIGN COMMERCIAL RESERVATION Principal Innovator: Stephan Richard Lee (ORCID: 0009-0008-1177-3857) Corporate Assignee: GenSi Holdings LLC // GenSI Technology and Infrastructure Services Official Operations Matrix: gensi.25.1@protonmail.com Timeline Horizon Priority: June 2026 Base...

Large Language Models for Research Quality Prediction: Applications and Empirical Evidence

DataNomics™ Vault App: Sovereign Multi-Agent Gated Repository, Automated Ingestion Countermeasures, and Cross-Thread Cognitive Synchronization Protocols [GenSI-Ref: DN-VAULT-2026-0920]

ABSTRACT, GLOBAL PRIOR-ART AXIS, & SOVEREIGN COMMERCIAL RESERVATION Principal Innovator: Stephan Richard Lee (ORCID: 0009-0008-1177-3857) Corporate Assignee: GenSi Holdings LLC // GenSI Technology and Infrastructure Services Official Operations Matrix: gensi.25.1@protonmail.com Timeline Horizon Priority: June 2026 Base...

The Extrapolation Crosses Zero Because the Heat-Capacity Difference Was Held Constant ── With a melting entropy of 10 and a heat-capacity difference of 30, the extrapolated entropy difference crosses zero at T=214.959393, or 0.716531 of the melting point ── let it vanish as the third law requires and it need not cross at all ── [Paper 503]

The Kauzmann paradox is spoken of as a property of supercooled liquids. This paper shows that crossing zero follows from holding the heat-capacity difference constant──held constant, however small, it always crosses. No new theorem or law is claimed. Scope of this paper (scope note): No new theorem or law is claimed──t...

Small Models, Big Claims: Choosing and Quantizing Local LLMs on an Integrated GPU

Running large language models locally on integrated graphics is now practical, but the community's model-selection advice is dominated by vibes: "just grab a 2B at Q4." This report presents a controlled, single-machine measurement study of twelve model/quantization configurations across three model families (Qwen2.5-1....

Window Theory • AI — Paper II: Landing and Verification

Window Theory - AI (wtAI): A Falsifiable Framework for Flexibility-Fragility Coupling in Large Language Models This is a six-part research series examining a structural pattern in decoder-based next-token-prediction systems deployed as dialogue and instruction agents. It is written for readers with a background in mach...

Comprehension as Compliance: How Grounded Content, Not Form, Moves Small Language Models to Act, and Where It Does Not

A small language model, given a task prefixed with a description of a recurring decision, will often recognize the decision and fail to act on it. This paper draws together a program of studies on that effect and states what drives it. The effect is carried by the content of the description, not by its form: a plain im...