Daily Update Thread from Dave #16
Replies: 1 comment 4 replies
-
Much of the discussion on GitHub centers on agents communicating with multiple agents. However, I believe we should start with the basics. We should first develop a functional agent builder, possibly using a tool provided by the tool maker. My suggestion is to adopt a machine learning approach. Specifically, we could create a tool maker that generates code, which in turn produces system messages to construct an agent capable of executing a task. The tool maker would then assess the agent's performance and refine the code generation process accordingly. This iterative cycle would continue for some time. This approach narrows our focus to a single issue: the development of the agent creation code. By doing so, we avoid the complexity of facilitating communication among multiple agents at this stage, since interactions would be one-to-one. |
Beta Was this translation helpful? Give feedback.
-
Check here daily updates.
Today's video: https://youtu.be/_iKIAlwNAN8?si=ljTwISVAc66tvAUi
2023-11-09 - Thursday
TLDR:
Beta Was this translation helpful? Give feedback.
All reactions