We might also consider that the LLMs will take traffic away from original sources, much like Google Search showing the relevant passages that answer your question. It's not always about outputting the exact passage, though this point may be tangential to the post & case at hand.
My primary usage of LLMs fit this pattern. Use Bard for simple programming questions, typically something I would have previously searched for via docs/reference/SO for. Now I don't have to filter the results or skim the pages for the right section, it's just given to me, and I can ask/make follow up questions or corrections. This here is what makes them much better than vanilla search, the ability to iterate and refine.
> We might also consider that the LLMs will take traffic away from original sources, much like Google Search showing the relevant passages that answer your question.
That said, Google has motivation not to over do it though as millions of websites are part of their ad network. If people stop visiting the sites, they lose money too.
Paradigms have changed, chat will be a primary modal for search in time. This is why Bing and Goog are both finding ways to get ads into our chat experiences. We might be forced away from ad dollars being the primary way to pay for the internet. With data privacy, ideally content based ad placement over individual targeting (think magazine style ads), the move to paid access to sites or content... there are trends that may combine to break the current foundations.
My primary usage of LLMs fit this pattern. Use Bard for simple programming questions, typically something I would have previously searched for via docs/reference/SO for. Now I don't have to filter the results or skim the pages for the right section, it's just given to me, and I can ask/make follow up questions or corrections. This here is what makes them much better than vanilla search, the ability to iterate and refine.