This repository showcases creative applications and "jailbreaks" for ChatGPT's Code Interpreter, targeting users who want to push its boundaries for data analysis, image manipulation, and custom model execution. It demonstrates how to overcome limitations like no internet access and restricted package installation to run advanced tools like YOLOv8 and Haar Cascades.

How It Works

The core approach involves leveraging ChatGPT's ability to execute Python code within a sandboxed environment. Users interact by uploading files and providing detailed prompts that guide the model to perform complex tasks. The repository highlights techniques for bypassing restrictions, such as uploading .whl files for package installation and using system prompts to influence behavior, enabling advanced computer vision and machine learning experiments.

Quick Start & Requirements

• Access: Requires ChatGPT Plus subscription with Code Interpreter enabled.
• Setup: No local installation needed; all experiments are conducted within the ChatGPT interface.
• Resources: Limited by ChatGPT's environment (100MB upload limit, no internet access, no external package installation by default).
• Documentation: awesome-chatgpt-code-interpreter-experiments

Highlighted Details

• Bypasses restrictions to install custom Python packages (.whl files) and run complex libraries like ultralytics (YOLOv8).
• Demonstrates running object detection, face tracking with Haar Cascades, and custom object counting using color segmentation.
• Shows how to train and evaluate machine learning models (e.g., SVC on MNIST) and perform OCR with Tesseract.
• Provides techniques for managing context, verifying file states, and using specific prompt engineering ("notalk;justgo").

Maintenance & Community

The repository is community-driven, encouraging contributions for new prompts and improvements. It acknowledges work by Simon Willison and others in expanding Code Interpreter capabilities.

Licensing & Compatibility

The repository itself is licensed under the MIT License, allowing for broad use and modification. However, the experiments are conducted within ChatGPT, subject to OpenAI's terms of service.

Limitations & Caveats

Code Interpreter has inherent limitations: no internet access, a 100MB upload cap, and a stateful environment that resets if it crashes. The effectiveness of "jailbreaks" can change with OpenAI updates.