chatGPT has a model only of language (a statistical model of what symbols are most likely to come next, given the symbol streams that it has seen).
It contains no other theory, knowledge base, or understanding of computational outcomes.
(This is why it's not even reliable for arithmetic)
So, while it can regurgitate or hallucinate, it can't aim for outcomes.
Getting there by prompt adjustment would be a long, iterative, and time-consuming shot, with only a dice-throw prospect of success.
To quote from a chatGPT3 response this morning – which clearly bears the mark of some "supervised learning" (to deflect potential liabilities):
As a language model, I am designed to generate text that is similar in structure and content to the text that I was trained on. While the text I generate may appear to convey information, it is important to keep in mind that I do not have an understanding of the concepts or ideas that I am expressing. I simply generate text based on patterns and associations that I have learned from the data that I was trained on.
Thanks Peter - can this be done in some programmatic/automated way? I am hoping to use this as a template for more advanced scripts; it would be much more helpful if I can render the HTML at the time the script is run.
Am I correct that when I want to use more modern methods of Javascript output such as those which use the canvas feature, I need to use a Custom HTML Prompt and create the canvas there?