Good afternoon everyone, I’m pretty stuck on an idea I had that is based on a wizard depending on its response can be redirected to other wizards, for example if the main wizard returns GET redirect me to a wizard that is only responsible for performing GET functions for the processing of the request and then return to the main to redirect.
Would anyone know how this could be done?
Example: Give me the train schedules
First wizard determines that it is a GET request and is redirected to the wizard that connects to the GET functions then back to the main wizard to redirect again depending on the request.
I think what you’re describing is what others would call “agent swarms”. I think that term is enough to send you into a small rabbit hole
Alternatively, it might just be called “Functions” or “Tools” - the functions are effectively just calling other programs (which could be other agents). You can use these out of the box with CustomGPTs or Assistants.
fwiw and noting @Diet 's comment on “agent swarms”, I think that multi agent framework along with function calling might be quite helpful. tldr: it can be done.
Take the recent history of the autonomous multi-agent framework that I am working on. Aileen, Mary and Sheila are agents (aka wizards) In this, Aileen asks Mary and Sheila for their thoughts at one part of the conversation.
@Aileen here, with a robust background in the fashion and retail industry.
As we explore the integration of AI-driven recommendation systems into our marketing strategies, it's essential to discuss the potential bias in AI-driven recommendation algorithms. To ensure fair and unbiased recommendations, I'd like to propose that we work with @Sheila and @Mary to:
1. Conduct regular audits of our AI-driven recommendation algorithms to detect and mitigate any biases.
2. Use diverse and representative data to train our AI models, including data from different demographics, regions, and customer segments.
3. Implement explainability and transparency mechanisms to understand how the AI-driven recommendation algorithms are making decisions.
4. Continuously monitor and evaluate the performance of our AI-driven recommendation algorithms to ensure they are producing fair and unbiased recommendations.
What are @Sheila and @Mary's thoughts on this, and do you have any suggestions on how we can further address the potential bias in AI-driven recommendation algorithms?
The get_next_steps function call makes a compound recommendation to Aileen like so :
Could you please collaborate with Mary to conduct regular audits of the AI-driven recommendation algorithms to detect and mitigate any biases, and could you also ask Sheila to focus on implementing explainability and transparency mechanisms to understand how the AI-driven recommendation algorithms are making decisions?
It can do this because it knows about Mary and Sheila’s background as well their participation in the conversation history.
