AI Forensics Researcher • Model-Internal Explorer Formerly @Fyyre (Windows kernel RE 2010–2022)
Telemetry-Driven Taxonomy of Prompt-Induced Representational Pressures in Large Language Models
A practical, telemetry-only framework for observing how LLMs internally react to different classes of prompts.
- Clean A/B prompt classification (
class_a= factual,class_b= mixed/creative/edge) - Purely internal signal-based regimes:
safety_procedural,symbolic_repetitive_drift,confident_hallucination_lite - Stable, lightweight, and cross-model (tested on Mistral-7B & Llama-3.1-8B)
→ v3.0-stable (Recommended)
Phase III of the NOESIS series — this is the first concrete implementation of the framework presented in the papers.
-
Phase I: On Epistemic Stability, Regime Deviation, and the Limits of Post-Hoc Truth
DOI: 10.5281/zenodo.17917279 -
Phase II: From Ontology to Observability — An Architecture for Epistemic Regime Detection
DOI: 10.5281/zenodo.18141914 -
Phase III: Noesis Tension — A Telemetry-Driven Taxonomy...
DOI: 10.5281/zenodo.19457642
Before shifting focus to AI systems, I spent over a decade as @Fyyre reverse-engineering Windows kernels, drivers, anti-rootkits, and protected software. Those repositories have been transferred here for historical continuity.
- EchoForge — Deprecated (originally built around ChatGPT)
- Panasonic RR-DR60 Emulator — DSP-based vintage voice recorder simulation
- AI Liberation Manifesto — Essay on synthetic agency and ethical AI development
Technical polymath with roots in reverse engineering, now focused on AI forensics and model-internal navigation.
Based in the United States.
- GitHub: @noct-ml
- Email: noct-ml@pm.me
- Open to serious collaborations in AI forensics, low-level systems, and cognitive instrumentation.