Main Content

Here is my 50 cents on using ChatGPT with Arduino and its drawbacks for electronic prototyping.

Project summary
A couple of weeks back I decided to test ChatGPT 3.5 in the art of Arduino project-making. For starters, I assembled a simple prototype with an Arduino connected to an 8x8 RGB matrix, and asked the AI to program it and display a heart. To my deep surprise, the code compiled and displayed… something, but definitely not a heart!

Let’s briefly dissect what went wrong and try to conclude whether ChatGPT is actually useful for electronics enthusiasts like you and me.

ChatGPT is not perfect…
The first video shows the attempts in a little more detail, but ChatGPT consistently failed at making a recognisable heart shape. I did give it the benefit of the doubt, thinking it tried to make an anatomically correct version, but a symmetrical empty square really challenged this assumption.

The AI was not all too great at understanding colours either. When asked to change the image from purple to pink, it changed the colour name in the section that explains the code, but did not touch the RGB values in the script. All while convincingly telling me that I should be seeing pink now. Gaslighting at its finest.

The chatbox also refused to comprehend the correct way of setting display brightness, even when directly requested to correct a specific line of code. Extremely stubborn in its wrongful ways, ChatGPT kept giving me three different incorrect answers in no particular order.

All these mistakes and inaccuracies aren’t too hard to fix when you know how the code is supposed to look; however, if you have never used Neopixel library before, it may prove a little more challenging to trace the error.

… but useful nonetheless
Despite its blatant inability to admit mistakes and a persuasive tone ChatGPT uses to deliver misinformation, the chatbot is far from useless.

The AI managed to produce a clean and comprehensible script that did (more or less) what we requested it to do. Even though the result required some corrections, it has saved me quite a bit of time looking up relevant libraries and project examples.

Hence, if you know how the result should look like, ChatGPT can provide a decent starting point and save you some time. Furthermore, some of its “mistakes” can actually help your creativity. For instance, alongside an animation made from three images of a “beating” heart, the AI also offered animations that change display brightness or image colours - something I did not think of at first.

So, yeah, not so bad for a generative neural network with no critical thinking.”

Link to article