Language model are actually fascinating, having this next word prediction able to fool real conversation is crazy.
But in the other hand, that's all what it is, prediction, and guessing (hallucination). I wish sometimes I get a 'I don't know' reply instead of indeed having always something to say.
As for MoI usage, I guess your best bet is to give example code of "similar" script to help the model to write the syntax correctly during your request.
I manage to do a bit of openCL for Houdini with chatgpt that way. At some point I wasn't sure who was assisting who, but I usually have my stuff done. It's kinda helpful for "non coder" people to get a bit of work done and get more autonomy. That similar task would probably take my co worker 5min to solve tho.
|