The term "hallucinations" is now commonly used for instances of AI making stuff up - like when I asked ChatGPT (before I had paid account) to recommend 5 books about a certain topic and two of the recommended books looked totally plausible, but when I tried to find them, I discovered there are no such books. This is where I see a big difference between GPT-3.5 and GPT-4.
>> I get a response that isn't grammatically correct/doesn't include punctuation
What punctuation? If you mean stuff like commas separating complex sentences, my English is definitely not good enough to spot that. But your mention of punctuation reminded me of problems that ChatGPT has with my native language... any chance you are using ChatGPT in a language other than English?