neomundi-lab/law-e-framework

https://huggingface.co/datasets/neomonde-lab/law-e-framework/resolve/main/README.md

Law E Framework – Thermodynamic Governance for AI Reliability

Short description

Law E is an operational framework that treats modern AI systems as thermodynamic information processes.
It introduces a native governance layer that observes the “energy cost” and coherence of model outputs, and uses this signal to regulate hallucinations and unstable behaviors.

This repository hosts the initial technical report describing the framework, its main equations and the first proof-of-concept design.

📄 PDF: Law_E_Framework.pdf

---

Why Law E?

• Large language models are powerful but prone to hallucinations.
• Current guardrails are mostly symbolic or heuristic.
• Law E proposes a physics-inspired governance layer:
  • monitors useless energy dissipation ΔE
  • tracks global organization / stability
  • regulates inference when the system drifts

The goal is to move toward self-regulated, energy-aware AI systems.

---

Current status

• Conceptual framework and equations defined.
• First regulator–selector POC in development.
• Next steps:
  • standardized hallucination evaluation (TruthfulQA, etc.)
  • CPU/energy proxy metrics
  • public demonstrator for selected models.

---

Contact & Collaboration

Created by Sébastien Favre-Lecca (Neomundi Lab)

• Website: https://neomonde.io
• Twitter:

If you are working on AI safety, energy-aware AI, or robotics and want to collaborate on Law E evaluation or implementation, feel free to reach out.

---

---

Experimental Reports (Zenodo mirror)

This repository also hosts a set of experimental research reports archived on Zenodo, documenting empirical observations and audit-level traces related to information–energy dynamics and endogenous regulation in artificial systems (Loi E).

### Available experimental reports

• Information–Energy Dynamics Simulator (Loi E) — Experimental Observations
  Systematic exploration of information–energy regimes in a minimal simulator, including non-uniform dissipation, distributed asymmetry, recoverability windows, and irreversibility thresholds.
  📄 PDF: Information-energy dynamics simulator Loi E — Experimental observations.pdf
  🔗 Zenodo DOI: https://doi.org/10.5281/zenodo.18483571

• Regulator–Selector (Reg-Sel): Measured effects of endogenous regulation in large language models
  Experimental evaluation of a minimal inference-time regulation mechanism applied to LLMs, without modifying model weights, prompts, or using external supervision.
  📄 PDF: Measured effects of endogenous regulation in large language models.pdf

• Filtre E — Audit trace example: information–energy regulation
  Audit-level illustration of explicit information–energy thresholding and regulated versus non-regulated response regimes.
  📄 PDF: Filtre E — audit trace example information-energy regulation.pdf

These documents form a coherent experimental foundation complementing the Law E framework. Interpretative discussions (e.g. negentropy, biological analogy, governance implications) are intentionally addressed in separate publications.

Further experimental reports and technical notes will be added to this repository as the Law E research progresses.