ChadGPT

ChadGPT aims to solve the problem of time spent and bounties paid out for low to medium-tier vulnerabilities.

ChadGPT catches many of these bugs, by allowing you to and will save you money on bounties and give you more resources to focus on higher-level vulnerabilities.

By creating a Chad front-end for ChatGPT that accepts GitHub repo links and allows the user to ask questions of the code, anyone can find vulnerabilities in smart contracts.

We used Filecoin Virtual Machine, LangChain.js, DeFi Hack Labs repo, and the ChatGPT API.

The core focus of the project involved integrating the DeFi Hack Labs repository into a database formatted for optimal training of the ChatGPT model. We employed the embedded training approach, which simulates an open-book examination, allowing ChatGPT to function as a student and enhance its ability to search and retrieve answers from the provided database. This method leads to improved predictive capabilities due to the pre-memorized information stored within the model.

In an upcoming upgrade, we plan to utilize the fine-tuning training technique to further augment the model's performance, transforming it into the more advanced GigaChadGPT.

To accomplish our objectives, we leveraged the Filecoin Virtual Machine for storing the custom vulnerability database and Langchain for training the model. We also implemented next.js to develop a sophisticated front-end interface for Chad.