[palata]

> This may be one reason ChatGPT et al feel so novel. They get rid of the noise.

I think I disagree. LLMs are much better at increasing the noise than they are at getting rid of it. Worse: they are really good at making it look like they get rid of the noise.

To me, LLMs have the potential to break the Web. Instead of search engines crawling the Web and allowing people to search its content, we may have to go back to trusting people: "I read this blog because I know who writes it". The day we can prove that LLMs were used to add tens of thousands of mistakes in Wikipedia (with a political agenda), will we still be able to trust Wikipedia the way we do today?

LLMs have the potential to systematically and automatically destroy the Web.

[bdw5204]

Wikipedia policy will hopefully prevent LLM generated garbage from corrupting it because some random web site is not a "reliable source" to add "facts" to articles. Today's Wikipedia is also remarkably effective at enforcing its policies. Most of its incorrect information scandals and notable long term vandal incidents happened in the 00s when it was still new.

That's the positive side of its policies. The negative side is that it often reflects the biases of its "reliable" sources meaning it often gets the same things wrong that the mainstream media and academia get wrong.

[technotarek]

I was being specific when I chose an LLM product like ChatGPT, not LLMs generically. I do so because it is a product like ChatGPT that boils away the noise to the end user, for better or worse. That is, to grossly simplify my point, it provides a single answer / response, instead of a long list of potential garbage for which I must dig, ponder, assess etc.