loud orbit
Would love some input on my project. I've only ever built for myself, so it may not be very user friendly but I'd like to make it more robust and friendly over time.
Note: Many of the existing experts use claude-v1.3 because I like its speed, capabilities, and price way more than I like GPT-4 atm. So you will want to change the model name parameter on some of the existing experts.
Check out my other project for an example of how experts can be implemented pretty easily into larger applications.
loud orbit
I have rewritten the prompting insights from the above repos into a simple to read and implement format. Please check it out here: https://github.com/Bradybry/chatXML/blob/main/chatXML.md
I also just uploaded the first in a series I am working on to quantitatively evaluate a variety of prompt engineering ideas using chatXML.
https://github.com/Bradybry/chatXML/blob/main/chatXML Evaluations/eval_test.ipynb
I'm most surprised that Claude beat all attempts by GPT-3.5 and GPT-4, while managing to do it with much lower latency and cost. When more people get their hands on Claude I suspect alot of people will breakup with OpenAI.
