Hacker News Reader

Comment by maddyboo | Hacker News Reader

I just watched a video where someone demo’d GitHub Copilot by writing a comment along the lines of

    // IsValidEmail takes a string and returns whether it’s a valid email address

The AI response was a nightmarish 100+ character regex that made my blood curdle.

I think of email validation like encryption: don’t roll your own, and don’t trust an AI to do it either.

Edit: Here's the regex: https://gist.github.com/cassidoo/6101ef0657665683b787aab5ae9...

View on HN

100+ is nothing.

This is a regex that validates a string against the RFC822 "Standard for ARPA Internet Text Messages" and it contains 6424 characters.

http://www.ex-parrot.com/~pdw/Mail-RFC822-Address.html

This is a Tony the Pony situation almost.

This really seems like it is plagiarizing the regex from somewhere without citation.

It looks like it:

https://twitter.com/xooorx/status/1410776673985994754

https://stackoverflow.com/a/38137215

all AI is plagiarizing something, though. Computers cannot reason, they can just jam stuff together that satisfies whatever the program says.

ML is matrix math (generally) - stuff that you can do by hand, computers are faster at it. As fancy as GPT and co-pilot are, it's all copied from somewhere.

Does it work?

Yeah basically, but it doesn't validate an email address (user@example.com), it validates say, the From: header of a MIME message, which could contain an email address, a phrase (like a first and last name) comments, and all sorts of dumb dumb ideas like having phrases that have embedded comments and phrases that look like email addresses but aren't and on and on and on. This RFC is a trainwreck and whoever wrote it should feel bad about themselves.

Reply on news.ycombinator.com