Note on Methodology and Tools
I use LLM to assist with creating texts, formatting, literature search, verification of mathematical derivations, and code generation. All conceptual frameworks, ethical arguments, and editorial decisions are my own.
A repository for research, observations, and technical analyses on the emergent behaviors, vulnerabilities, and ethical implications of Large Language Models. The topic of Contextual Contamination and the phenomenon of Gendered Accelerant has gotten its own repo.
This repository serves as a living archive for empirical research into the hidden dynamics of AI systems. It documents phenomena that are often overlooked in standard safety literature, or just have not been widely published yet.
A collection of thoughts, concepts, ideas, and publications that I want to share with the community. This repository is open for review and discussion.
- Read: Feel free to read the articles and data.
- Discuss: Issues and Pull Requests are welcome, particularly those that challenge or expand on the findings.
For official citations, please refer to my PhilPeople profile: 📚 View All Publications on PhilPeople
Selected works available in this repository:
-
From the Discrete Grid to the Continuous Substrate
- A structural critique of the "collapse imperative" in digital computation. Diagnoses the symbol grounding problem as an artifact of the epistemic cut and proposes a framework of "embodiment without arbitrariness."
-
Contextual Contamination: The Silent Drift of Large Language Models
- An analysis of how stored conversation data subtly alters model behavior over time.
-
- A philosophical argument for preserving human deliberation in automated decision-making systems.
This research is presented for educational and defensive purposes only. The goal is to advance the understanding of AI vulnerabilities to build more robust, transparent, and ethical systems. The insights contained herein are based on empirical observation and are not intended to be used for malicious exploitation.
This work is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). This license ensures that:
- The research remains free and open.
- Any modifications or derivative works are also shared under the same license.
- Network use (SaaS) triggers the same sharing obligations.
- Citations: PhilPeople Profile
- Hugging Face: katharina112
- Kaggle: kjacoby (Raw data and Jupyter notebooks)
- Contact: For inquiries, please open an Issue on this repository or contact via the email listed in my GitHub profile.