Artificial Intelligence

I'm sorry, I'm currently in the process of setting up the Zed editor, since that one claims to have been designed with AI in mind. But I doubt that will teach electronics to ChatGPT or AIstudio. I already provided it with all the relevant snippets from Espressif's Programming Guide (via the webinterface), so it knows how to access the correct peripherals. I even hand-picked which one to use and discussed with the AI, why that is. Still the resulting code isn't combining it with the other specific requirements I have, but more some straightforward stepper motor example code. But that's not what I want. And I'm not really sure what do do here. Do I need to feed it the hundreds of pages about the silicone as well, and some good books on electronics and microcontroller programming and why and how we use the peripherals on them? (And combining them was the specific task I wanted to use AI for, so I can't feed that in, or I'd do everything myself.)

And what I've also seen it do is bullshit me with the memory allocation. It will sometimes home in on C++ programming. But more how we do it on computers. I specifically instruct it to avoid dynamic allocation, but it's super generous with everything and do it nonetheless. Then I tell it it the data structure is on the heap, and it says sorry and it's going to fix that... Changes around the data structure and it again ends up on the heap. But I can't use it that way for a library. This will work on an ESP32, but not on other microcontrollers. And you can't just be very generous with everything on resource-constrained systems like a microcontroller.

And in the third example I tried to make it do some more involved maths, with the same stepper motors I got from a broken 3D printer. I thought I'd build a robot arm and have AI do the kinematics. And then also the inverse kinematics and use vectors for that. And it's been a while, so likely things have improved a bit since a few months ago. But back then it failed miserably at doing the maths. It could recite the Wikipedia article on what inverse kinematics is. But that was pretty much it. It had zero abilities to apply or understand the maths behind it... It was however capable of coding a nice 3D JavaScript visualization in the browser on how it failed and the angles and positions would look like in that imaginary robot arm. I had to assist a bit and fix several mistakes. But that was something that worked.

I'm not really sure what you're trying to tell me. Prompting these AI tools is a bit hard. And I frequently make mistakes. But I'm willing to learn. It's just that my experience shows it's somewhere between copying code from Stackoverflow like we used to do. And AI has the added benefit of being able to tie it into things and at the same time also write the boiler plate code around it. But I don't think it's super clever. I just see it struggle a lot with any more advanced programming concept, maths, electronics... That doesn't mean it's useless or can't do other (easier) things. But I thought it had ingested some books on programming already and I don't need to teach it for example what dynamic memory allocation is, and why we avoid it at times? But without that knowledge you just can't go far from the usual blink LEDs and beginner examples. And I'm not sure whether the editor or webinterface is at fault here.

Edit: I'm pretty sure just downvoting me won't get a discussion going. We might not share the same perspective, that's why I gave an outline of mine. But are you interested in learning something, or do you just want to push your uninformed opinion?