Hacker News Reader

The problem for Google seems hard to me. They scrape original content then find ways to present the best parts of it without sending you to the content creators site.

Now they will have a harder time finding human generated content. So AI will start training itself on AI generated data. Which sounds like a loop that leads downward in quality.

View on HN

This assertion regarding quality sounds like an appealing but probably wrong guess, like curse of dimensionality.

There's other related issues too, though. Like, the incentive to publish "good stuff" goes down, as there's much less quid-pro-quo traffic from Google to reward the work. You're right in that I'm guessing, but Google does seem to be headed down a road of eating itself.

https://www.cl.cam.ac.uk/~is410/Papers/dementia_arxiv.pdf

As another commenter here said a while back, pre-2021 web crawls will be like the low background steel of the web.

"Definition 3.1 (Model Dementia). Model Dementia is a degenerative process affecting generations of learned generative models, where generated data end up polluting the training set of the next generation of models; being trained on polluted data, they then mis-perceive reality. "

Ah, it has a name even, so I'm at least not alone in thinking it's an issue.

Reply on news.ycombinator.com