Hacker News Reader

Comment by 2000UltraDeluxe | Hacker News Reader

2000UltraDeluxe·

But how can you trust it to provide accurate information?

When I've played around with Bing, I've been seeing hallucinations and outright false data pop up quite regularly.

My initial assessment of LLVMs is that they can be great writing aids, but I fail to see how I can trust them for search, when I can't use it for simpler tasks without getting served outright falsehoods.

View on HN

You have to follow the citations. They have information; the headline result doesn't tell you anything except "here's where we think you should look". That's a search problem.

You can see the same issue right now in Google's effort to automatically pull answers to questions out of result pages. Frequently it gets those answers wrong.

But that’s not how humans function. They won’t follow citations because it’s added work. Nine times out of ten, they will take what the AI spits out at face value and move on. Also those citations have a higher probability of being created by AI now as well.

Yes, and humans will just skim search results rather than actually read the article. Or trust the Wikipedia page or book. Or believe the talking head. When they are not invested in the answer. But the occasions where it matters, we do read the article and/or check multiple sources and decide if it is bullshit or not. I don't much care if I get the wrong answer about how many tattoos Angelina Joli has, but I find myself comparing multiple cooking recipes and discarding about half before I even go shopping.

Humans differ. If the information is not controversial most often accept it and move on. If the information is controversial most handwave, but a large group further checks and if they arrive different conclusions they start to get active on replacing incorrect info.

Reply on news.ycombinator.com