https://github.com/anon0411/g-series-alignment-pathways - G-Series Alignment Pathways: Structure-Based Interpretability without Intent Insertion

This release includes a full set of structural alignment experiments extending the G-series framework, focusing on whether repeated circuit patterns can yield externally interpretable alignment even under perturbations. The repository contains:

• G-Bx-P, G-Ex-P, and G-Fx-P experiments
• Structure flow diagram and summary documentation
• Reproducible Jupyter notebooks for each experiment

No explicit intent was inserted; instead, structural recurrence and feedback were used to test interpretability.

본 릴리즈는 G시리즈 실험군의 구조 정렬 실험을 확장한 자료로, 의도 삽입 없이도 구조 반복과 피드백만으로 외부 해석자가 정렬 신호를 인지할 수 있는지를 검증합니다.

- Jupyter Notebook
Published by anon0411 10 months ago