Crap... this thing knows how to build out RegularExpressions

yeah, maybe. But that particular regex has to be on the internet at least tens of thousands of times.

Not that hard for the AI to learn it from stack overflow or similar without connecting any dots.

googling "regex to match a mac address" actually gives you a better result with a better explanation.

"([0-9A-Fa-f]{2}[:-]){5}([0-9A-Fa-f]{2})"

which is shorter, not much harder to understand, works for both : and - separators, and the guy even explains that IEEE802 allows both : and - as valid symbols.

so yeah. If you were following the AI you can get a decent working result. But if you were implementing some sort of mac matcher that you'd want to use on user data or all kinds of data, you wouldn't be following IEEE802 specs and miss a bunch of stuff.

So there's that aspect to this.

@Hazarth Sadly that is how Copilot worked.

ChatGPT has only trained on a pre-defined set of data : I was trained on a dataset consisting of a wide variety of text from the internet. This includes a diverse range of content such as books, articles, websites, and more. The training dataset was carefully selected and cleaned by my creators at OpenAI to ensure that I could learn to generate high-quality, coherent text.

So not sure how far this was going, but for me, it managed to create a script I could use in lua for RedM (read dead server)

Maybe based on the context, but it is pretty amazing.