Yea, that's no exaggeration. The hardest part is continually remind yourself to keep trying new things to push it further.

I asked GitHub Copilot to write a couple of Python functions, and I was pretty impressed. I dont mean, "write a function to add two numbers" or "parse some text". I defined an Nx3 and an Mx3 Numpy matrices (a & b), told it (in comments) that a & b were two arrays of 3D points. I asked it to write a very fast function to compute all distances between pts in a and b, and return it. It did it immediately and the results were correct, so it's a start.

I had already written my own function to do it with vectorized numpy math (no slow loops), and optimized. I wanted to know which was faster.
The result is always a symmetric MxN matrix and diagonals are zero, so I knew my outer-difference MxN matrix had to be wasteful.

I benchmarked CoPilot's code against my own: their code was 3X faster in large matrix tests.

The second example: I told it I have a laser with 700 nm wavelength. I gave it some specs, like the diameter of the laser, an aperture size, and told CoPilot to write a function to compute and plot the laser image projection on a plane that was X mm away. It did it first try. It looks something like this image:

Yea... amazing is just the start.