Prusa Core One HT – Finally, it's done!  

RE:

Nice! Now that you have done it, could it be ported to 6.4.0 (and future firmware versions) easily? Any specific reason why you did not start based on 6.4.0? For the Core One it seems to be a reliable firmware, apart from the bootloader blue screen.

As a side note, did you get ChatGPT to actually point you to the relevant files and functions in the source code and suggest what to change? I have tried this a couple of times -- with ChatGPT and Grok so far, using registered but free accounts -- and was quite disappointed: They just kept directing me to the Github repository in general, rather than searching it and suggesting specific files and function or variable definitions. I was wondering whether there are access limitations (technical or contractual) on Github which prevent the LLMs from doing that? 

RE:

Nice! Now that you have done it, could it be ported to 6.4.0 (and future firmware versions) easily?

Hello Jürgen. I've already thought about that. I assume that upgrading to version 6.4.0 is no longer an obstacle.

Any specific reason why you did not start based on 6.4.0? For the Core One it seems to be a reliable firmware, apart from the bootloader blue screen.

As with the Prusa XL, I was also not satisfied with 6.4.0 on the C1. I only used version 6.3.4 because I felt I had more success with phase stepping when searching for noise in the Y-axis than with the current firmware.

As a side note, did you get ChatGPT to actually point you to the relevant files and functions in the source code and suggest what to change? I have tried this a couple of times -- with ChatGPT and Grok so far, using registered but free accounts -- and was quite disappointed: They just kept directing me to the Github repository in general, rather than searching it and suggesting specific files and function or variable definitions. I was wondering whether there are access limitations (technical or contractual) on Github which prevent the LLMs from doing that? 

I have been using the paid version of ChatGPT for quite some time. At some point, I couldn't get any further with the free models. In the beginning, there was only one mission: Chat GPT was supposed to “guide” me so that I could understand for myself what I had to do to get functional firmware.

However, it quickly became apparent that my lack of programming knowledge and incorrectly phrased questions to ChatGPT led to a dead end. That changed when I added the link from Github from metacollin and wrote that I only wanted the PT1000 sensor evaluation in the Prusa firmware 6.3.4. And then it started, and ChatGPT made a 180-degree turn toward the finish line.

I was impressed by the evaluation of the failed attempts to create the firmware on the PC. For every error that appeared in the console, a change or improvement was made after I told ChatGPT what error messages appeared after the last command.

Without ChatGPT, it would have taken me months or even years to understand how Prusa's source code works. But I must also say that if you have learned programming languages and know how to write programs, it is safer and easier than using AI.

Unfortunately, I don't have that mindset, and AI was the missing piece of the puzzle to get me to my goal.

RE: Prusa Core One HT – Finally, it's done!

I've done it a few times now, but not for firmware, but to add a few QOL improvements to Prusaslicer. (inner/outer/inner wall order, and option to define the ironing angle)
I find the best at this sort of stuff by far is Claude.
The quickest route is to work out yourself exactly what files hold the code that you will be adding to or changing. Then describe what you're trying to achieve in your prompt, and provide the raw url links to the files that need altering.

I was amazed by how quick and painless it was to achieve the desired result...I know the basics of many programming languages, but this (with C++) was way beyond my pay grade.

RE: Prusa Core One HT – Finally, it's done!

Is that a bambulab build plate?

RE: Prusa Core One HT – Finally, it's done!

Posted by: @gb160

I've done it a few times now, but not for firmware, but to add a few QOL improvements to Prusaslicer. (inner/outer/inner wall order, and option to define the ironing angle)
I find the best at this sort of stuff by far is Claude.
The quickest route is to work out yourself exactly what files hold the code that you will be adding to or changing. Then describe what you're trying to achieve in your prompt, and provide the raw url links to the files that need altering.

I was amazed by how quick and painless it was to achieve the desired result...I know the basics of many programming languages, but this (with C++) was way beyond my pay grade.

Indeed, Claude seems to be ahead of the competition when it comes to programming. My wife uses Claude Code for work. I have not talked myself into any paid AI subscription for hobby purposes yet, but when I do so, it would probably be for Claude as well. (I also like Anthropic's corporate approach to AI -- they openly discuss the risks and how to mitigate them, where other vendors seem to focus on the hype.) 

I would hope that getting a handle on a large existing code base -- e.g. scanning all source files to find the relevant definitions for some intended change -- should be well within the capabilities of an LLM. Seems even easier than coding? Architecture of complex, entirely new systems is a different matter, but that is not what we need here. 

RE: Prusa Core One HT – Finally, it's done!

Yeah it's hands down the best.
The infuriating thing is I have free enterprise access to ChatGPT, but honestly its not great for coding...its great simple coding, and other things but my primary use for AI these days is coding thats way beyond what I could achieve with my own knowledge/skills, so I basically have to exhaust my cap with Claude, then he kicks me out for six hours, then I go back to it.

I cant emphasise the time I've wasted with ChatGPT (going round in circles trying to achieve a simple task), and I've concluded I'm better off waiting for Claude to let me back in.

A simple trick I've learnt with Claude is to start your session with him around 5 hours before you really want to get busy with it just give him a vague outline of what you're planning to do, then after 5 hours hammer him until the cap expires...then when he kicks you out you'll only have to wait around a hour or so before the cap resets. Its a PITA, but that way you can get virtually double the cap per session (I hope that makes sense)

RE: Prusa Core One HT – Finally, it's done!

Cool... I'll give Claude a try. I hadn't heard of him until today.

RE: Prusa Core One HT – Finally, it's done!

I'm a professional Java/Python/C++ developer and until AI our primary resource is/was StackOverflow. I might get stuck on say how to write a mocked unit tests, go onto Stack Overflow and maybe find an example that works. StackOverflow is a place where you ask a question: "How do I do X?" and you get answers where the participants can upvote or downvote the answers. Even if I get an answer to a similar question,  at a minimum I would need to modify it get it to do what I needed. Still, that would require having a deep understanding how the code in question functions. 

Recently they've rolled out Claude at work. I've been using it to generate documentation or some of the other aspects of coding which are tedious like writing unit tests. We've had mediocre developers leaning on it heavily to the point where they don't understand the code that was generated. Sometimes this can be amusing. Example Claude spat out:

if (a != b) || (a == b) {
}

Which is funny because it's a tautology, i.e. it's always true. The developer didn't think to question this.