My son once published a book on debugging. It contained 50 programs, 10 each in 5 different programming languages, each supposed to contain exactly one bug that student/reader was expected to find. There was a lot of text about debugging techniques and the back of the book contained the solutions. (Over the years, users found 3 additional bugs, but let’s ignore that.) A couple years ago, he asked an AI bot to find the bugs in the ten programs written in C++. It found exactly one of them. Last week he repeated the exercise with ChatGPT. It found all ten!
the progress is incredible … just x-link that to the “Half-marathon winning robots” … and just project that out 3-5 years or so …
Exciting times, in which we live … indeed,
(and the fact, that nobody really knows how all this works - makes everything so much worse)
I’ve been a computer programmer for more than 30 years.
In fewer than 5 years, the LLMs went from being worthless to being much better than I could ever hope to be. I’m honestly not sure why I still have a salary 
It’s finding and fixing dozens of our bugs every week. Just last week it fixed some ancient half-decade-old hidden issue and improved performance by like 1000x.
PS GPT 5.5 is supposed to be another big upgrade over last week’s GPT 5.1. Not only are they getting better, they are getting better faster than anyone can keep up with anymore.
I’ve gone from not using it to using it extensively over the last three months. They is knowing what outcome you want, and that’s where my 35+ years of experience still comes into play. I care more about the destination and less about the journey.