ChatGPT is just the brand. I guess you're using GPT-4, but if you're using the default model (GPT-3.5) that would certainly explain below-expectation results.
Mystery solved! The answer to your OP is that the reason anyone can trust ChatGPT for code is that they use a much better model than the one you're using! GPT-3.5 is ancient and way behind GPT-4. In fact, there are now tens of organisations who have developed model classes way ahead of GPT-3.5.
chrisjj|1 year ago
"ChatGPT" is a model, says its provider.
But yes, the web page says model GPT-3.5.
calderknight|1 year ago