I've tried a few of my favorite ChatGPT 3.5 & 4 prompts, and I am getting terrible results compared to GPT 4. Claude 2 is hallucinating on almost every response, giving wrong answers, and saying things like:
> The problem is I am rounding the approximation before casting it to u32. This will round the value 83.33 to 83, instead of the expected 120.
It's pretty good at maintaining superficial coherency, but the content feels slightly weaker than GPT 3.5.
Poetry writing seems to be stuck, so far, on GPT 3.5-level doggerel, even when prompting with instructions that cause GPT 4 to write fairly good poetry.
EDIT: With some prompt tweaking, I've gotten some vaguely competitive results for poetry. Still not really in the same league as GPT 4.
> The problem is I am rounding the approximation before casting it to u32. This will round the value 83.33 to 83, instead of the expected 120.
It's pretty good at maintaining superficial coherency, but the content feels slightly weaker than GPT 3.5.
Poetry writing seems to be stuck, so far, on GPT 3.5-level doggerel, even when prompting with instructions that cause GPT 4 to write fairly good poetry.
EDIT: With some prompt tweaking, I've gotten some vaguely competitive results for poetry. Still not really in the same league as GPT 4.