This is happening in
#VLSI space as well. The advancement of
#FPGA technology has only exacerbated it, as many engineers are using it as a crutch to avoid thorough simulation. They are too rushed or lazy to understand the benefits of constrained random verification.
#CRV
There's an epidemic of people building software that doesn't work.
Testing has become a lost art. We are now in the era of "impressive demos," and that's all that matters.
I was raised with an old-school mentality:
1. By default, we assumed software didn't work
2. We proved it did work by writing tests
3. We only delivered when those tests were successful
I come from a time when computers always did what we wanted them to do. That's what coding has always been about!
This has changed with AI models:
When you use a Large Language Model as part of your application, you can't ensure the outcomes are the ones you'd expect. These models are unreliable. They are hard to control.
This makes testing and evaluation more important than ever.
I think there will be a reckoning. Some of the demoware today will go nowhere, and people will stop trusting software. (I already know folks who don't use anything that's marketed as "AI.")
We need to get back to the fundamental principles of working software.
Testing must be a first-class citizen.