Recent decline in ChatGPT 4.0 responses to coding prompts

[thinktank]

Hey mate, this sounds like an instance of ChatGPT becoming “lazy.”

@razvan.i.savin

• They are lazy because we do not give them a direction → we must be specific enough, or "it will chase its own tail" if we are not specific enough, it will use its values to generate something that is in its bubble. Simply put → add more knowledge and specific instructions.
• They are lazy when there is a limit on context length under the hood (this is set to avoid so much waste of computation made by the user. I still think that on the API we should have control of this, and if it is above the limit, in a range set by the user, to pay extra).

Check out this discussion.

My suggestion, if you’re using ChatGPT Plus, is to make a CustomGPT that has your exact specifications.