No need for AI slop.
18 months ago I would have agreed with you; case-in-point, when I asked ChatGPT to make some changes to a simple Arduino sketch back then, it made a real hash of it. All of that has now changed. Completely.
I currently have it manipulating in excess of 30,000 lines of Python code for my control panel; the way it keeps track of deep, complex interactivity is nothing short of mind-blowing. And the code it's writing is absolute textbook. But it is quite common for it to overlook something (albeit only the tiniest fraction of what I would have overlooked) - thus - it's a team effort; it writes the code and I make sure the code runs how I want it to. Rinse & Repeat.
With a paid ChatGPT subscription comes access to Codex - their code-writing specialty AI; it's next (several) level(s). Case-in-point, I added live-view onto my control panel; doing it via screenshots was too slow - doing it via Siglent's streaming functionality had too many issues to be reliable (even with WAV ACC on) - but their web functionality isn't too bad - so we took to emulating a browser and capturing and displaying the web streaming directly. We didn't know what some of the required parameters were initially so I had Codex literally take control of the scope (after asking permission) on the LAN and it figured it out by itself (writing it's own test programs on the fly to do it) - then detailed a report for ChatGPT to impliment - and it's working flawlesly.
Days of "sloppy AI code" appear to be long gone if my experience is anything to go by; what it's written for the control panel developed into something extremely complex under the hood with things like separate command / verify pipelines (similar to what modern aircraft use) - operating asynchronously with Windows HID events (so there's not a lag on the display whilst commands/status are communicated with the scope) - working around bugs (like how the scope reports that a setup recall is complete when the scope's file system is still needs a couple more seconds to catch up). I don't even bother trying to write Arduino code anymore; initial AI attempts were embarrasing but now what it produces is rock solid and is produced in a fraction of the time.
Completely different game now.
I already know I am going to get a lot of hate for my opinion here. So let me play the devil's advocate no less!
I must say, I would not even classify this software as "AI Slop", it looks like a moderately-polished first prototype. I have seen the profound impact firsthand, where the integration of AI into day-to-day workflows in my own job has helped speed up productivity, and churns out continuously good software in my experience. This doesn't excuse the human element entirely, however. The user needs to be adept enough to know what they wish to produce, but AI does a pretty good job of a first-pass, regardless of the user. It is becoming more adept with specific skills, and needs to be studied a bit closer, but looks quite promising.
I have had Cursor generate some applications for me already that run in-brower, powered by a Python back-end. I must say for a single-time use application, it heavily speeds up my workload with device bringup, PI testing, etc... While I can specific test procedures and bringup processes manually for example, I do not have the time to code an application from the ground up with so many complex features. In this day and age, AI is not nearly powerful enough to take over my own job, but it is highly useful by me and my team on a day-to-day basis for certain tasks, and helps streamline things. I would argue that the quoted post is more a demo of an app than an ad for GPT services. For people only beginning in the field of EE (and similar), do not become reliant, as the few first years build the foundation for most careers; but that is not to say don't not use it to help with taskload.
I will
not, however, comment on TestController, as when I posted a previous thread on some software I'd developed
myself to integrate alongside the tool, I could tell it was assumed the project was AI generated, I was met with what I could only interpret as "flack".
Home
Help
Search
Login
Register
