|
One big difference between googling and asking an LLM is that it tends to be much easier to judge the legitimacy of information from a google result through informal heuristics. This is because google just links to a website, and there are all sorts of other factors that you use to subconsciously rank how reliable information on a particular website is, that mostly have nothing to do with the raw information. For example, we all have different weights for how much trust information we get from, e.g, Wikipedia, stackoverflow, high profile bloggers in your domain, conspiracy websites, listacles, blog-spam sites etc, marketing materal etc. If the info is on a social site, is highly upvoted? Is the info from an academic journal, arxiv.org, an academic blog, or a wordoc you downloaded from scribd [1]? Yes, this is judging a book by it's cover, but it's a heuristic that tends to work well. By contrast, LLMs present all information to you with the same confidence in the same homogeneous interface. There is no external context. So all your normal heuristics for judging reliability are broken.
- [1] https://xkcd.com/1301/ |