I would like to see this expanded, I think it's a bit unfair to assess its abilities with so few examples. My hypothesis is that a rosetta stone with a thousand examples with a vector database hooked up to it so you don't hit the 32k token context limit would lead to much better performance.