This video is going to demonstrate how to use multiple LLM chat systems to collaborate on a coding and scripting request. The method is to give basic instruction to each system and have them respond to each other’s output. All that’s needed from the user is to copy and paste each response.
Multiple rounds of testing have confirmed that this conversation remains coherent between the completely different chat bots and that the end product is workable, manageable, and actually better. More testing is needed to confirm limitations and benefits, but it can create and even convert code for the same function to completely different code bases.
In this example, GPT4 and Claude AI are used to generate a conversation. The initial prompt is detailed but not too long, going over the basic parameters of how to respond to each other and denotes which chat system is responding to the other.
The example involves creating a PowerShell script that creates a new user on the PC. The script prompts the user for any necessary information and gives the option to make the user an admin. It also mentions that additional functionality can be added or configured within the script by bouncing ideas off each other.
The conversation continues by feeding the entire output, including the initial request, into Claude with a small change to indicate the second user responding. Claude refers to the initial script given by GPT and includes other ideas for adding them to other user groups.
The revised script is then fed back to GPT to see if it has any issues or suggestions. The collaborative discussion between the chat systems results in a comprehensive script that maintains readability.
The final script is outputted and reviewed by the human intermediary. It is then fed back to Claude to check for any syntax or coding issues. The corrected script is agreed upon, and the full script is outputted.
After testing, it is confirmed that the script includes all the elements discussed and performs various tasks, including drive mapping and home folder location. The video concludes, and further testing will be done on the script.
