While making a kernel, the instructor was using particular GPT for defining the kernel, Why did he use a ChatGPT to define these kernels

While the instructor was making kernels, once with open ai and then with an open source, I notice instructor particular selecting the GPTs.

If there are different GPT for different language models, why can’t while making a kernel, one use a whole GPT language model and let the kernel choose the most effective or closely related GPT automatically based on the feed given to the program?

Like When before anything was asked GPT, it used to give any feed related to the work fed in the program. But ChatGPT revised it create a better algorithm in case of getting a better analysis of input and output(which I still feel needs to improve)