Deep Seek chat made a custom P1 MAX7219 driver
TonySpiner
Posts: 61
Playing with the free Deep Seek chat bot this morning, took several iterations but the result works.
TonySpiner
Posts: 61
Playing with the free Deep Seek chat bot this morning, took several iterations but the result works.
Comments
object attached
That looks pretty good. Deep Seek seems to know what it's doing and the code is very readable. This seems like a good way to learn the P1.
Where did all the gear come from? That's not just programming there.
the gear is a couple of very simple PCBs I designed with KiCAD and had made by JLCPCB. Some bits & bobs from Ali express, including RGB LED rings and I²C LCD, all hooked-up to an old P1 project board. Two input PCBS are daisy-chained to provides 32-bit serial data for encoder and button patterns, other PCB is a 16 x 7-seg LED display with 8 LED indicators, again serial data from the P1. There is also a MIDI output port, still on some strip-board hiding in the background.
That doesn't sound cheap. What was the assembly cost?
The PCBs are approx £5 each inc. shipping. No assembly cost, all through-hole parts, easily hand soldered, cost pennies. Can't remember the cost of the RGB LED rings or I²C display but they were not at all expenisive, check Ali-express etc. I already had the P1 project board, it's at least 10 years old and was originally used in a long-forgotten camera trigger project! Best of all KiCAD and Spin Tools are free!
I think this is all VERY low cost, it's kept me busy for a few months!
Learning P1 SPIN coding, alongside using AI to develop P1 code, seems to be working for me. The first time I fixed some Claude AI generated code I realised how much I was learning and that AI assist could be an effective way of working for me.
Oh, you did the assembling. Sorry, misread.
Yeah, I can't figure out if AI will be mostly used to learn, or used to do the whole job and relieve the person of needing to know much. I'd like to think the prior, but I can imagine a future where all chips are just seen as various means to some end, with each chip having some strengths over others, but they all are more-or-less putty to AI. I enjoy learning and having a familiar platform to build on, myself.
Works well in the first case. Great at regurgitating boilerplate code scraped from many good open sources. Like a search engine.
Is terrible in the second case. Pure vibe coding always makes a mess, even when there is good boilerplate code to reference. Trying to get it all glued together grows more chaotic with each addition. It bloats badly too.
It has been demonstrated that attempting self-teaching LLM models rapidly degrade. They only work at all when provided with well written sources.
PS: And that doesn't deal with the horrible levels of electricity needed to run such Smile.
Chip, you'll be happy to know "mostly used to learn" is the ultimate result for me after going hard at it with multiple models for multiple years. @"Stephen Moraco" helped tremendously by getting the documentation together and MCP built for agents and that has gotten it to a level that encouraged me to keep going.... and to keep going with the Spins (meaning 1 and 2, not that I was getting drunk... well, maybe drunk with power that I could wield with the help of ai models
). Frankly, I spent my first couple years noodling around with generic languages on P2 until I really began to grasp the power in the architecture and facilities in P2 that make it unique - namely the XBYTE system.... no more layers of cycle-wasting and memory-wasting shims beneath my Lisp/Prolog for traceable logic inference engines OR my continuous neural-net statistical machine learning machine controller with smartpin i/o OR SQL b-tree string interning storage OR smartpin/virtual function data stream memory/buffer manager without the need for garbage collection.
any and all of those, the ai models were able to produce functional enough implementations that I could see them work, with some confidence, after very heavy scrutiny. it was that very heavy scrutiny that taught me both the algorithms and the syntax of these systems.... and at the end I could recognize where the implementation was weak and by then I understood how to fix it syntactically in Spin2. the ai fall down when trying to pull these together into proper objects that work together in a system. when using AI alone, it tends to fix what is in front of it, and forgets how things have to work in a complete system.
why do I say all this now?... mostly because last night some clarity came to me, that caused me to get up and throw out %80 of the code written by ai, around the foundation of the my system WAY TOO MUCH MORTAR (cruft) BETWEEN THE BRICKS (xbyte). and I swear I will demo this stuff on monthly meetup and video RSN (real soon now)!
@refaQtor, that all sounds excellent. Keep going!
I find AI really useful for showing me the shape of things I don't know about and for springing me out of potholes.