Hacker News Reader

I think that as long as actual human beings continue to contribute content, AI would eventually "refine" that out and incorporate that into newer models. Even if say, 95% of the content in the dataset is AI generated, it will still capture the remainder.

Of course, I don't know if scaling AI with more data will work to improve it or not, and I'm sure that anyone really knows right now.

View on HN

We can introduce healthy cows into the downer cow population. Even if say, 95% of the herd is exposed to BSE, it will still capture the healthy parts of cow carcasses we feed the downer cows.

(Which is to say... I think there are more than homeopathic concentrations of questionable training data. Or at least I believe that was the assertion of the OP.)

Reply on news.ycombinator.com