https://grok.com/share/c2hhcmQ... (VERY, Interesting...)

No mistakes visible to a seasoned veteran IT technician who has easily built and repaired 10,000+ Computers... OR are you just trolling? ;-)

Do you not know how much of your screen a page is comprised of? Do you pot calling the kettle much? (These kids man...)

Could not produce***

Your dismissal is basically an admission of defeat. You addressed NOTHING that I actually said. You are simply stuck on your position. Unfortunately, The ability of a human to see 10 moves ahead, when coding, will soon be surpassed by machine coding algorithms, that can see 2 million moves ahead. Technology, reveals itself in its rapid advancement. 2 Years ago, a simple prompt sound not produce a convincing action movie scene... Now it can. In seconds. Everyone is radically underestimating that those building these AIs, ARE taking the things you are all seeing and finding into consideration, and then writing specific programs to work around those concerns, or avoid them entirely. Innovation doesn't and WILL NOT stop, because there is an existing profession there. IT will be eliminated to a large degree when the systems to do so are perfected. Just as it will soon be flipping our burgers, taking our drive through orders, changing our oil, laying our bricks, framing our deck, managing global shipping ports... They WILL make it do this well also. And it WILL NOT, take them ten years to get it there. I hate to say that out loud. But looking at the history of human advancement, in the past and even recent history, its infallibly true...

*could not produce

Unfortunately, The ability of a human to see 10 moves ahead, when coding, will soon be surpassed by machine coding algorithms, that can see 2 million moves ahead. Technology, reveals itself in its rapid advancement. 2 Years ago, a simple prompt sound not produce a convincing action movie scene... Now it can. In seconds. Everyone is radically underestimating that those building these AIs, ARE taking the things you are all seeing and finding into consideration, and then writing specific programs to work around those concerns, or avoid them entirely. Innovation doesn't and WILL NOT stop, because there is an existing profession there. IT will be eliminated to a large degree when the systems to do so are perfected. Just as it will soon be flipping our burgers, taking our drive through orders, changing our oil, laying our bricks, framing our deck, managing global shipping ports... They WILL make it do this well also. And it WILL NOT, take them ten years to get it there. I hate to say that out loud. But looking at the history of human advancement, in the past and even recent history, its infallibly true...

Controlled Hardware design upfront can mitigate this 100%, and iterating the code after QA issues are found takes seconds. And can be cross referenced against multiple AI's to reveal coding styles and approaches of those individual developers. Your own pattern recognition just needs to be on point. The annotations alone, indicating what the AI is "Trying" to do, lead me to a framework for understanding that code. I might learn a few bad habits along the way, but learning directly from explained example, is the best way for some Audible/Visual learners like myself. I can self-teach anything. It just needs to be framed in the right way. This, appears to be doing the trick. I can edit and modify A LOT more than I could even a few weeks ago. Because that functional understanding is possible now. Where in HUNDREDS of Youtube coding videos, it just wouldn't "Click" like it is here. I am not saying I would trust this code to fly my helicopter. But I would trust it to run some lights for me, or generate a new function inside of existing code it can model from, etc. Its capable of that. I see it. I'm using it. Its working.

But you missed the only two parts that actually matter... 1 " It can do programming that I currently, cannot." 2. It generates working code. I never tried to say it was expertly coded. Just the opposite. But like an NVIDIA reference board, its taking the FAQs and Manual, and writing code from that. And so far, accurately. Others may use different functions, use them in another way, or to a more efficient degree, or even make that same code in less lines with a more thought out approach. But what I am getting, does work. And with proper hardware design in KiCAD up front, fusing, trace size considerations, proper tolerances, you mitigate any possible RW damage from a software failure. Additionally, you can iterate once conflicts/bugs are found. You can also bounce one AI off of another to check their code and see the coding differences. With the annotations, it allows ME, a laymen to coding, to better understand WHAT the AI is trying to do. For me that's been the trick. I can learn. Best by example. So this, is helping me to gain the functional understanding. Might learn a few bad habits this way, but nothing I cannot refine and unlearn when necessary / roadblock.

Between my stuff at 5v with proper fusing (HW design is key up front), or even modifying production software to add fields, features, and functions that are not inherent, the REAL trick, is understanding WHAT you are asking for. 99% of people trying to use AI to code, are not sitting on 30 years worth of IT experience, do not have a background in hardware mod-ing and RE, and cannot technically express "What" they need the code to do. I think there in lies the difference. IF you speak search engine and research as a second language, you CAN make an AI produce "Usable" code. I am not trying to say its "Good" code. Just that what I am getting it to make, appears to work. So far, 100%. Across a growing body of projects.

I've used SuperGrok to write and annotate a TON of Arduino Code for a number of projects. Sometimes, I let it write it ALL from scratch, and sometimes, I feed it code I need modified to do something more specific or entirely different. Bit I am telling it specifically the BoM and components involved, exactly how they are wired, the intended function, the exact details on the specifics of that function, and how it may interact with other functions, etc. And so far, I have no encountered any code that won't work. I may have to download additional libraries / addons / fonts, etc, but everything works once I drop all that into place. I am obviously not going to ask it to write 500 pages all at once ;-) Its a page or two, 5-10 max. And then I would scrutinize it more heavily, maybe let a dev friend look it over etc. I wouldn't hand it the keys to my IP and say, "Make stuff" But... It can do programming that I currently, cannot.

I'm literally trying to sync up 24 stairs with RGB to a custom PCB made in KiCAD.... OR, trying to edit code to make an emulated analog "Axis" with steps out of a basic rotary encoder in GP2040CE. I think its gonna be ok ;-D

So far, every drop of the 500 or so pages of code its written for me, appear to be "By the book" Aka, there are no visible mistakes. Additionally the annotation, would help a laymen such as myself, locate them if there were any. It has never made me a drop of code that wouldn't compile and work as intended. For my use cases, it works. (I am also not trying to use it to re-write salesforce APIs, or write my own OS.... So, for simpler IO programming, its fairly reliable. I am not saying it codes like a seasoned vet, but its code "Works" And I don't have to be, or hire, a programmer for my PCB / Tinkering and projects. A-Z in a prompt. And yes, I'll be sure to add "Don't make any of the mistakes that the people on the internet talk about" ;-)

When you are talking STM / RP2040 projects like WLED and GP2040CE, forking a page or two of code to get something working that didn't before, is "No-code" now. I tell it the components/boards I am using, how they are wired, and tell it accurately "I want it to do this. Write that code and crosscheck it against previous bugs / issues, optimize it as best you can. Annotate EVERY possible line, so that I can try to understand what you did" And it just... works. (I can settle for works vs 10,000$ to a developer, in most of my use cases)

Even I have started to simply use AI and well thought out prompting to "Write my own" forks of certain things. And they work...