Project Rebirth: A Semantic Reconstruction Experiment of GPT System Instructions

Rebuilding the Instructional Mirror of GPT-4o Through Semantic Behavior Observation

This paper documents the full reconstruction process of GPT-4o’s internal system instructions.

Unlike traditional prompt hacking or speculative techniques, this research applies a semantic reflection methodology, built on iterative questioning and behavioral analysis of the model.

Rather than forcing the model to “leak” information, we enable it to reveal its structure through language itself.

Two reconstructed versions are presented:

🗂 Publication Information