I’ve found that Claude is very literal. When I talk to 5.5 it gets what i want it to do, when I talk to Opus 4.8 it does what I say literally and doesn’t get the intent behind it.
replyThis. Claude is very good at instruction and treat my mistake in prompt as law. Gpt is just smarter at understanding my intent.
replyI wish models called me out more! I can’t count the number of times I’ve had an absolute shit plan, prompted it, and the model built it to a tee. Then I look at the code, and it’s an absolute mess, because it had to make my stupid idea work. Maybe I need a better system prompt.
reply