#wired article

This is about an interview with Jingna Zhang, founder of Cara. Cara is a platform for artists, posting art, where their art is safe from Generative AI. In the interview Jingna talks about Cara, what it stands for, what SHE stands for, and why artists are mad about Generative AI. It's a really good article and Jingna Zhang has the most eloquent statements. I'm starting to worship that woman.

I think this should explain enough to the people who think artists are just being whiny and salty about AI stealing their jobs. Generative AI could have so much potential if artists' copyrights were being respected and acknowledged, which is not happening right now. Please, read and share this article!

Some excerpts:

If somebody asked me how many bugs there are in my Heart library, the honest answer would be that I don't know. Currently the library's GitHub issue tracker shows no open issues. The library is about 21000 lines of code (excluding blank/comment lines), and most of it has never been systematically tested. An optimistic estimate would be 1 bug per 100 lines of code, so about 200 bugs. (A more realistic estimate might be 400-500 bugs.)

How can I justify publishing a piece of software with hundreds of bugs? Well, mainly because I think it satisfies a need. I use it every day. And when I find a bug, I fix it promptly. I haven't encountered a bug in the library since January, and the bug in question was discovered by code inspection, not testing. To me, Heart seems like good, reliable software.

How is it possible for a library with hundreds of bugs to appear reliable? Well, I started writing Heart in 2017, and it hasn't changed much in the past year, so the commonly used codepaths have been heavily tested. While it's entirely possible for software to exhibit bugs that pass unnoticed by the user, it's likely most of the remaining bugs occur in unusual situations.

A user other than myself might use the Heart library differently, exercising code I rarely (or never) use. To them, Heart might not seem so reliable.

Now according to the article below (citing Wired magazine from 2012), a modern high-end car "has" about 10 million lines of code.

It's unclear whether all that software runs on the car itself, or if that number includes software used to design and test the car. During the 1990s, Microsoft applications sold to the public averaged about 5 defects per 1000 lines of code, which was remarkably good IMHO. I can't imagine the software quality of a Lexus is any better than that. So a modern car probably has at least 50000 software defects.

It's mind boggling.

Even if all that code is actually running on the car, it doesn't mean the car is unsafe or unreliable. The Wired article hints that most of that code is for entertainment, not safety or reliability. Still, there are certainly bugs in every large software system, and the more you deviate from the conditions under which the software was tested, the more likely you are to encounter them.