Live data from Hacker News

Generative AI could make search harder to trust

wired.com

51–60 of 255 posts

Re: Generative AI could make search harder to trust

#51
post #31

I have to say--the opening paragraph doesn't describe a reality I'm familiar with: >Web search is such a routine part of daily life that it’s easy to forget how marvelous it is. Type into a little text box and a complex array of technologies—vast data centers, ravenous web crawlers, and stacks of algorithms that poke and parse a query—spring into action to serve you a simple set of relevant results. Web search has, f…

Sounds like a success if that means people see more ads while trying to find what they actually searched for.

Nationalize Google.

Nothing will change as long as search is optimized for revenue over user value.

Re: Generative AI could make search harder to trust

#52

I wonder if there will be a human information/knowledge equivalent of low-background steel (pre-WWII/nukes). Data from before a certain point won't be 'contaminated' with LLM stuff, but it'll be everywhere after that. https://en.wikipedia.org/wiki/Low-background_steel

People are vastly everestimating how unique this problem of hallucinations is.

It seems to me it relies mostly on discounting just how much we've already had to deal with this same problem in humans over the millenia.

The problem of proliferation of bad information might be getting worse, but this isn't native to generative AI. The entire informational ecosystem has to deal with this. GPTs compound the issue, but as far as I can tell, no where near what social media has forced us to deal with.

Re: Generative AI could make search harder to trust

#53
post #35

Earlier quoted context omitted.

I experience that when I try to google for technical problems I'm having at work, but otherwise searches still go pretty well for me. I just had to google a bunch of races that I wanted to run. The top result was always the event's own web site. When I google some news, relevant news articles always come up. The last search I did was for how to display a ket vector in LaTeX. The top result was the StackExchange artic…

Google search to retrieve specific factual information is pretty good. Google search to retrieve anything opinion related has been horrible and infested with blogspam for years (hence people searching Reddit to get that kind of info).

Really? I've been finding it doesn't even find stuff it used to in certain documentation (I'm talking like things it found maybe a year ago), "searching in quotes for this stuff", things that other search engines (bing, kagi) are indexing just fine - And since I've switched to using these engines more when I'm searching things for programming work, it's definitely been a lot more helpful than google which often just seems to be missing a ton now

Re: Generative AI could make search harder to trust

#54

it's almost like AI just repeats data its fed on, even incorrect data, without any real intelligence to determine if the data is correct.... /s Its not simply garbage in garbage out. There is no logic to verify and analyze the data. You are simply told what is popular in the data.

All of a sudden the saying "eat your own dog food" takes a twist and is no longer fun.

Re: Generative AI could make search harder to trust

#55

More amusing and frightening is when people search about themselves and turn up AI generated crap. Googling yourself was always a lucky grab bag, with the possibility of long-forgotten embarrassments being dragged up. But at least you'd have to face facts. Now I hear of people discovering they're in prison, married to random people they've never met, or are actually already dead. What is this going to do to recon on…

This is a new way to be anonymous too. Someone post something true but nasty about you? Have LLMs cook up dozens of preposterous stories - you're secretly a rodeo clown, you write childrens books, you built a castle in Rome, you once drank a goldfish, etc. Increase noise to drown signal.

This is essentially the service Reuptation.com claims to provide. Jon Ronson's "So You've Been Publicly Shamed" describes the site games SEO to flood the search results of controversial figures with banal nothing posts[1]. The difference being that actual humans had to create that content.

In the near future, the web could become opaque with LLM schlock, but at least it may grant people a right to be forgotten.

[1]https://www.businessinsider.com/lindsey-stone--so-youve-been...

Re: Generative AI could make search harder to trust

#56
post #35

Earlier quoted context omitted.

I experience that when I try to google for technical problems I'm having at work, but otherwise searches still go pretty well for me. I just had to google a bunch of races that I wanted to run. The top result was always the event's own web site. When I google some news, relevant news articles always come up. The last search I did was for how to display a ket vector in LaTeX. The top result was the StackExchange artic…

Google search to retrieve specific factual information is pretty good. Google search to retrieve anything opinion related has been horrible and infested with blogspam for years (hence people searching Reddit to get that kind of info).

I suppose it never occurs to me to search for opinions. I'm not even sure how I'd got about it, even if search weren't broken. Blogspam is what I'd expect to see.

I'm more likely to start at a place that aggregates reviews and try to hallucinate which ones were written by people who know what they're talking about. That usually seems to work.

I imagine that somewhere out there is a person who bought the product and reviewed it on their blog or made some enthusiastic social media post about it, and that's what you'd want to locate were it not for the spam. But I don't expect any search engine to be able to find it for me.

Re: Generative AI could make search harder to trust

#57

I wonder if there will be a human information/knowledge equivalent of low-background steel (pre-WWII/nukes). Data from before a certain point won't be 'contaminated' with LLM stuff, but it'll be everywhere after that. https://en.wikipedia.org/wiki/Low-background_steel

People are vastly everestimating how unique this problem of hallucinations is. It seems to me it relies mostly on discounting just how much we've already had to deal with this same problem in humans over the millenia. The problem of proliferation of bad information might be getting worse, but this isn't native to generative AI. The entire informational ecosystem has to deal with this. GPTs compound the issue, but as…

The thing is when you call a human on bullshit, they usually can't back it up well enough to pass the smell test. When you call an AI on bullshit it can instantly fabricate plausible, credible seeming sources/evidence.

A human's lie is different than an AI's hallucination, since it's still based on (distorting) the truth, whereas the hallucination is based on an invented reality (yes I know it's applied statistics and there's no true model of the world in there, but it can report as if there is)

Re: Generative AI could make search harder to trust

#58
post #27

Earlier quoted context omitted.

This was how the "internet" functions in the book "Ender's Game". There is a small sub-plot about how he had to give a fake persona credibility on the untrusted network in order to be able to leverage a creating a fake account on the trusted network.

I find the xkcd interpretation more realistic: https://xkcd.com/635/ > Explained: https://www.explainxkcd.com/wiki/index.php/635:_Locke_and_De... >

I love that interpretation, but in today's retweet driven world of politically commentary, I actually find it quite plausible that pseudonymous kids with no grasp of the real world who think rational political debate is the nonsensical slogans they're spouting on the internet become major Twitter influencers that actual politicians want to court for their "authenticity" and "willingness to say the unsayable", and maybe their dank memes.

Re: Generative AI could make search harder to trust

#59

Recently an article came out where someone said that the company I work for is a big user of WebAssembly, but the reality is that we don't use it. After finding the contributed article (on a well-known news site, not Wired though), it looks like a tech founder might've been using ChatGPT to write an article about the uses for WASM. The arguments were generally sound, but I don't think that anyone did the work to manu…

This is kind of like the advent of spellcheck, where a whole class of errors started to appear regularly in almost every article because publishers stopped paying for the human labor to manually review for things like homonym or word ordering errors. Except much worse, because it could allow spurious or even harmful facts to accrue and spread instead of just grammatical mistakes.

Re: Generative AI could make search harder to trust

#60

More amusing and frightening is when people search about themselves and turn up AI generated crap. Googling yourself was always a lucky grab bag, with the possibility of long-forgotten embarrassments being dragged up. But at least you'd have to face facts. Now I hear of people discovering they're in prison, married to random people they've never met, or are actually already dead. What is this going to do to recon on…

This is a new way to be anonymous too. Someone post something true but nasty about you? Have LLMs cook up dozens of preposterous stories - you're secretly a rodeo clown, you write childrens books, you built a castle in Rome, you once drank a goldfish, etc. Increase noise to drown signal.

I think Boris Johnson tried that by saying out of the blue: he makes model busses. There was some thinking at the time that he didn't want the brexit bus to show up in searches and was trying to game search results..

I don't think it worked.

Post reply on HN