i am guessing that’s “does not fit your use case“ i have not had any problem with it in the software i have used that rely on it. i am not saying AI is not capable of fixing bugs, as mentioned i use it for a lot of things my self. but even Fable 5 hallucinate and have suggested wrong and stupid things to me that would work just fine but be a bad solution to a problem.
AI will solve your bug and make your problem go away, its excellent at brute forcing and can work on a problem for days if you let it, but its the solution that usually is the problem, in many cases its sub optimal, and will quickly build up technical debt.
lets say you want to enter a room but the door i locked, AI might suggest to remove the lock that way you can go in and out of the room without any issues, while a owner of the door would say you need to get the key first, the lock is there for a reason. now if this was a more abstract concept where the door was some kind of complex logic that required deeper knowledge of the system most people would agree/trust the AI with whatever assessment it made. it would solve your problem and you would be non the wiser, but it would not have solved the problem in a way most people would agree is the proper way to solve it.
Anyway i just stopped by to give my 2 cents and my experience with AI tools, i don’t feel like arguing for or against different views. Everyone is entitle to their opinion and mine is that AI tools generate too much sub optimal and bad code for me to use it in production code.
its still fun to play around with, and i will continue to do so until the start charging the same as the API prices. no idea what the actual cost of this would have been, but i would never have been willing to pay the API price for this
Oh, no. llama.cpp was not working correctly. It was the issue I linked.
I don’t think I have really seen it actually execute on something like this recently. It sometimes suggests things that are really stupid so I can pick between them, but it usually feels like it was grasping at straws to come up with 2 / 4 suggested options.
Like, do you want to do… (1), the actual fix, (2), delete the repo and jump out the window, (3), another possible but poor solution, (4), DDoS Github, until the code spontaneously fixes itself?
Which is why you need to really bug it on the solution. Or just let it go, ask it to find a good solution and look the other way.
I usually will present it an issue and if it’s complicated then I will ask it what it thinks would work as a possible solution and how we would implement it.
If you’re not happy with it’s answer then go back and forth until it fits. That’s why I like the Superpowers plugin. It usually has at least a mostly decent idea and then asks extensive plan questions to refine it.
Absolutely, and it is greatly appreciated. I have been wondering what actual developers have felt for a while since to me it has done everything I have asked and more and can now basically one shot entire tools and features.
That’s probably only part of the picture, but I think it would easily be in the $10-15k range just based on the Fable 5 usage.
I’m mostly using Opus 5.5 now. Fable is too heavy on the API limit’s and doesn’t really seem any more capable. And when you count that it would have a screeching fit literally every time I asked it to do work it was completely unusable garbage for me.
Opus 5.5 seems to be decent, haven’t seen any big mistakes yet, just the occasional hallucination now and then. I mostly use 5.5 for text work and docs. Fables text/docs are imo just hard to read and poorly structured, it just don’t convey information in the same easy to understand way that Opus 5.5 does.
Some of this could just come down to English being my second language, and that fable use words and sentence structuring that i am just not used to, but i think a lot of people agree that it does not make the greatest texts. I am hoping for a Fable 5.5 that perform the same (or better) but also writes text the way opus 5.5 does.
(i am emphasizing 5.5 here as opus 5.0 was just bad xD )
People who are qualified to review your code just don’t want to review AI-written code for free. I don’t really even think you are necessarily getting a poor response, in fact, I think the response to this post has been relatively glowing so far, all things considered.
It is not a reasonable ask to request others review your AI-written code, that you don’t understand, for free. Even if it’s amazing code….. The developers who have the prerequisite knowledge of the systems you ask about and are capable of reviewing code in large batches with extreme detail are few and far between and also stretched thin. They don’t have the time for silly tasks like this and most of them are so busy that they are looking for any down time to be spent on doing literally anything but development.
I don’t mean any disrespect. I think going about solving our own issues like this is admirable compared to doing nothing, but I think you just need some help understanding where other people’s head are at when you ask this sort of thing.
