roles.chat

LLMs like ChatGPTs are powerful. They leverage prompt instructions to act as a specific role to respond you.

What if several different roles chat together to do your job…

For example, when you require to install a software, there may be 3 AI ChatRoles involved:

• WebSearcher: be good at choosing search keywords and picking proper webpage results
• Downloader: knows how to download files
• ShellExecutor: writes/executes shell commands, to make it works

Note that, there may be several sub-conversations between these roles. For example, download url may be broken, shell execution may encounter exceptions. In these situations, roles may require more web searches to get it right.

This is what we want to achieve: to build a huge specialized AI ChatRoles repo:

• prompts based roles
• fine-tuned specilized model roles
• task coordinator roles
• action roles, like web surfer, API invoker, shell executor, etc. (with security concerns).
• big roles: several roles bounded together based on routines
• …

You may create/verify/refine/report, (and most important)chat with each one of the ChatRoles in the repo on the fly, right beside your chatbox!

Behind all these scenarios, we utilize the vectorStore/LLMs, to match the exact ChatRoles to your conversation.

More over, letting ChatRoles to have sub-conversations, has the advantage to control the context length, as the big boss, you needn’t know about each details, and things are settled!

First things first, we are defining the specifications for ChatRoles and coordinations!

So, stay tuned.