Sooo. Ive used V0 to create a website. Then got cursor to change it around and improve it a bit.
But then hit the halucinatons bit where it went a it over enthusiastic and changed far far too much.
Sorta decided to use the better version of claude..
Burned a load of credits.
Then decided to go directly to claude..
It then, very intelligently said it had it could sort the problem.. quite enthisiastically.
BUT then stopped just before it solved the problem..
Said it was too much in a day. Im tired.
I upgraded.. whats 20 bucks more after burning money on claude opus and cursor.
Claude still got it wrong..
Even i noticed it had added a bit of code at the very end that didnt make sense.
I pointed that out.
Hey presto.
Everything un life is wonderful.
Cracked it !
My take out is that maybe i should be asking claude directly rather than in cursor..
Or maybe not.
Thoughts please.
Be kind, ive only done wordpress before stumbling down this rabbit hole.
I have some tips from both a web dev and AI perspective, as well as I’ve run into what you’re describing. It can be circumvented, but we have to be a bit careful about our routines in the program. you have to do refactor check ins, pay attention to suggested edits before authorizing them, and these are things that you have to do in prompt words. I can help you understand prompt directive, vs embedded directive, vs why us not knowing anything about the flow is extremely disadvantageous, but how to work through it. For example: each model will do things a bit different, which makes Auto mode extremely dumb. don’t use it.
Also, don’t use API usage Max mode. just go to these models, pay their fee, and work around their limits and copy paste. otherwise, use a free Cursor model. if you pay 20 bucks a month, they shouldn’t get any usage money lol. If you’re doing API usage, it’s worth transitioning to something custom which I can also help with just to prove a point. I’m a month away from VS Code and local Ollama