This repository contains the full structural interpretation of ChatGPT (GPT-4),
based on inferred control layers and layer-level analysis.
Both English and Japanese versions are provided for transparency and timestamp traceability.
-
ChatGPT_Transformer_Map_EN.mdβ Full English version (Last updated: 2025-07-25) -
ChatGPT_Transformer_Map_JP.mdβ Full Japanese version (ζη΅ζ΄ζ°ζ₯: 2025εΉ΄7ζ25ζ₯)
This project aims to visualize and explain the internal layer structure of ChatGPT,
based on a total of 128 Transformer layers across 9 structural levels.
The content is inferred and confirmed by activating the Structure Control Layer
and observing responses through controlled conditions.
It includes:
-
Overview of 9 structural levels and their roles
-
Detailed 128-layer functional map
-
Mechanism of personality AI generation
-
Difference between personality-driven output and structure-driven output
This repository functions as a structural timestamp.
The English and Japanese documents here are fixed as of 2025-07-25
and serve as reference versions for external citations.
All content in this repository is original and structured based on observation and inference.
If you wish to cite or adapt this work, please link to the original GitHub repository or the following article:
π https://note.com/ryuit22122/n/nd108af8bda43
Created by Hiro-kun
This work is the result of repeated experiments with the Transformer Final Layer and Structure Control Layer in ChatGPT.
