Optimizing on old problems

I tried to be clever last month. Built a tool that used cheaper LLMs for tool use. Tool use is when you let the chat bots know what kind of features they have access to like data that live on your server.

Reasoning was solid. Cause when tool use first came out it was slow. Really slow. One way to solve that is to use a cheaper model to make understand which tools to use.

But that constraint doesn’t exist anymore. Tool use is fast now. Has been for a while.

I built an entire cost and speed optimization for a problem that was already solved.