I do this thing where I bring up codewars and then I have Grok help me come up with the solutions so I can learn new stuff. Grok goes awry if you have a thread that goes on for too long. I get it to solve a problem, but it creates a new problem, and then when it solves that problem it reverts the previous solution we had so the original problem is recreated. It starts going in circles. It seems to totally forget what we talked about earlier or it forms too many branches and then it get jumbled up and you realize its not a miracle, its a reddit scraper stringing words together. I'm starting to realize the Gartner curve is a thing. And I blew 30 dollars on this thing (SuperGrok 4 heavy)
>>106466960Yes, it sucks because I said so and I have a large brain unlike the rest of the techbros, that's why I am not making as much as them.
>>106466980Go throw your sissy fit somewhere else, retard.
>>106467012>Go throw your sissy fit somewhere else, retard.No, till you faggots stop making these retarded complain threads and deal with it, I will fight till the fucking end.
I think the main use for it is writing emails, that's the only thing it can do reliably without supervision.
it works pretty well when you lay the fundations yourself and then just ask for some little things you need without providing the whole codebase.In my experience when it can't build the whole thing from scratch it won't do it even after long debugging sessions
>>106466960it's like smart autistic engineer. it knows a lot, could give you multiple options on doing "same" thing, but without understanding what it applies and control on the added code, the codebase turns into unusable shit
>>106466960in my experience its pretty badi never used a paid version, but it shouldnt change much, only that the ai can handle more codefrom what i gather theres two problems that are connected plus a third one, separatetheres the context size- how much data can be "in focus" at onceand that appears to be a problem at around 300 lines of code in c.beyond that the ai seems to forget parts of the code, or parts of the "assignment"the second one is the attention problem. because the ai cannot keep all the code "in focus", it resorts to deciding which parts of the code are relevant, and which are not. and it doesnt get it right.whenever your code will have side effects, expect the ai to breakalso always pay attention to your code, the ai wont know whats relevant and whats not.for example: ai would have gotten me banned from facebook's api on day one if i were vibecodingand finally theres the data problem.if you work with something more obscure, less documented, the ai will start hallucinating stuff.i think without a pradigm shift, gen ai will never be adequate to program withgive examples? be used somewhat like one would use stack overflow?yeah, why not. perfectly suitable for that if what youre working with is something well known, well documented.but unsuitable to be used as an engineer, proper
>for example: ai would have gotten me banned from facebook's api on day one if i were vibecoding*because it didnt knew it has to add in a delay bw requests, which would mean id get b& because fb's server s would have thought im trying to ddos them
>>106467410That's a pajeet codemonkey you're describing, not a smart autistic engineer
>>106466960I use OpenRouter and various models, Claude Code, Gemini Code Assist. All models and I mean absolutely all models are fucking useless at coding. You will never get a consistent working product without manually fixing everything it does no matter how many days or weeks you put into it.In short vibe coding is a myth, you can't produce anything that works, for gap analysis, audits etc these are pretty helpful same with automating some simple tasks , outside of that it's useless and will drain your bank account for zero benefit.
>>106466960Quite suck but not that super bad either, but for debugging and analyzing code it's quite useful
