Hacker News Reader

> Will AI always be better? I thought it gets worse the more it eats its own output.

This isn't a real issue, there's just a large cohort of people who want to think it'll go away on its own. They like saying things like this because it's a very convenient belief for them.

View on HN

Likewise there's a large cohort of people who pretend diminishing returns don't exist since their investments depend on that not being the case.

Also the lack of factoring in the current cultural backlash.

Lots of free data was available because nobody cared or knew it was being used as training data a few years ago.

But now? Why publish a side project on GitHub if the current marketing for AI is effectively “we will use this to eliminate your job”?

I’ve noticed an increased interest in keeping side projects private in my circles.

> This isn't a real issue

You've just tried to sweep away a massive concern without justifying it

The entire advance in reasoning models comes from training on other AI output. So does Claude's "constitutional AI". The concern is entirely made up.

Reply on news.ycombinator.com