Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

ChatGPT Search

openai.com

Post reply on HN
Print view View original

901–910 of 1001 posts

Re: ChatGPT Search

#901

Post by FuriouslyAdrift »

marcus_holmes wrote: post #651
freediver wrote: post #241

Been thinking about this a lot [1]. Will this fundamentally change how people find and access information? How do you create an experience so compelling that it replaces the current paradigm? The future promised in Star Trek and even Apple's Knowledge Navigator [2] from 1987 still feels distant. In those visions, users simply asked questions and received reliable answers - nobody had to fact-check the answers ever. C…

I agree that this is the core question, but I'd put it as: Who Gets To Decide What Is True? With a search paradigm this wasn't an issue as much, because the answers were presented as "here's a bunch of websites that appear to deal with the question you asked". It was then up to the reader to decide which of those sites they wanted to visit, and therefore which viewpoints they got to see. With an LLM answering the que…

Non-citizens cannot vote in FEDERAL elections, but some states allow them to vote in LOCAL elections.

There are few more restrictions here: https://www.usa.gov/who-can-vote

Re: ChatGPT Search

#902

Post by marckrn »

gkfasdfasdf wrote: post #893
marckrn wrote: post #295

OpenAI also provides a Chrome Extension ( https://chromewebstore.google.com/detail/chatgpt-search/ejcf... ) to trigger a search from the omnibar as the default search engine. If you don't like that (like I do), you can also manually add it under Site Search using https://chatgpt.com/?q=%s&hints=search

How did you find the extension? Searching the Chrome Web Store for 'SearchGPT' turns up a ton of third-party extensions, not made by OpenAI. Also, there doesn't appear to be a way to search by developer.

They presented the link in the ChatGPT interface. Maybe this was an A/B-test thing, idk.

Re: ChatGPT Search

#903

Post by azemetre »

coffeemug wrote: post #507
whizzter wrote: post #300

This is probably Google's Altavista moment, by making their results crappier by the year in search of Ad dollars everyone has felt that there is a potential for search to be better and once that becomes available they'll be in a continious game of catch-up. Yes, Google has their own AI divisions, tons of money and SEO is to blame for part of their crappiness. But they've also _explicitly_ focused on ad-dollars over a…

Google is entirely to blame. It would be trivial for them to train a model to rank sites on a scale of SEO garbage to nobel laureate essay, then filter out the bottom 50%.

Why would Google do this when they make billions from SEO garbage? Those sites are serving ads after all.

Re: ChatGPT Search

#905

Post by netcan »

mcpar-land wrote: post #855
GolfPopper wrote: post #602

Every time I see one of these topics, I go ask chat GPT a question to which I know the answer on a topic where I would like to be able to get useful answers to similar questions that I do not know the answer to. This time it was, "Did Paul Edwin Zimmer write a fourth Dark Border novel?" (Real answer: Yes, Ingulf the Mad . You can find the answer on his Wikipedia page.[1]) ChatGPT's[2] answer: "Yes, Paul Edwin Zimmer…

A lot of these responses are so funny. "Um, when _I_ asked this question to ChatGPT it gave the right answer!" Yeah, because it's flaky! It doesn't give the same answer every time! That's the worst possible outcome!

Funny, yes. But... a webforum is a good place for these back-and-forths.

I'm on the other side of this fence to you. I agree that the conclusion here is that "it is flaky." Disagree about what that means.

As LLMs progress, 10% accuracy becomes 50% accuracy. That becomes 80% accuracy and from there to usable accuracy... whatever that is per case. Not every "better than random" seed grows into high accuracy features, but many do. It's never clear where accuracy ceilings are, and high reliability applications may be distant but... sufficient accuracy is not necessarily very high for many applications.

Meanwhile, the "accurate-for-me" fix is usually to use the appropriate model, prompt or such. Well... these are exactly the kind of optimizations that can be implemented in a UI like "LLM search."

I'm expecting "LLMs eat search." They don't have to "solve truth." They just have to be better and faster than search, with fewer ads.

Re: ChatGPT Search

#906

Post by mppm »

Abacus AI has had a similar feature for quite a while: When you ask something it searches the web and uses an LLM to analyze and aggregate results. Though that sounds useful in theory, in practice search results are such garbage nowadays that this very markedly decreases the quality of the answers. I have now turned off the search feature completely and only use LLMs for their internal distilled knowledge.

Re: ChatGPT Search

#907

Post by boplicity »

yalogin wrote: post #886

The main reason for google search’s success is it drove traffic to websites and both had their share of the ad pie. With ChatGPT they would take a much much bigger share of the pie and so if it becomes successful in replacing google search’s UI, it will fully replace the extreme distributed nature of the internet and make the traditional/big content providers even more powerful. I am sure content creators will find a…

This is a huge issue, actually. Google has many, many organizations on its side, simply due to the mutually beneficial relationship they have. That's something that will likely benefit Google in the short and medium term, though I can still very much see how this threatens them in the long run.

Re: ChatGPT Search

#908

Post by topaz0 »

sethammons wrote: post #801
topaz0 wrote: post #438

Earlier quoted context omitted.

Now we get to see how OpenAI will do similar harm, more effectively and at much greater environmental cost.

I feel like we have an opportunity to break a feedback loop. SEO worked because of links. No links in my chatgpt discussions. Ah, but what about all the junk ai gen content you may counter, but that stuff only works because of SEO links. As more people abandon searching for links for discussions, the SEO usefulness diminishes. Maybe to the point where many parasites stop making SEO shite in the first place.

But to your points. I think the problem with your analysis is that it forgets that the real driver of the junk is the advertising environment. The SEO links were profitable because the advertisers were willing to pay a few cents for space on those pages. Yes, the incentives are changing for the teenagers who are churning out text and adding seo links to their stable of cheap websites, but the advertisers are going to find a way to manipulate consumers that's compatible with the new order. I don't know what that will be, but whatever it is will depend on information pollution just as much as the current one.

Re: ChatGPT Search

#909

Post by HarHarVeryFunny »

FuriouslyAdrift wrote: post #901
marcus_holmes wrote: post #651

Earlier quoted context omitted.

I agree that this is the core question, but I'd put it as: Who Gets To Decide What Is True? With a search paradigm this wasn't an issue as much, because the answers were presented as "here's a bunch of websites that appear to deal with the question you asked". It was then up to the reader to decide which of those sites they wanted to visit, and therefore which viewpoints they got to see. With an LLM answering the que…

Non-citizens cannot vote in FEDERAL elections, but some states allow them to vote in LOCAL elections. There are few more restrictions here: https://www.usa.gov/who-can-vote

It'd make sense if paying taxes gave you the right to vote.

Taxation without representation and all that.

Re: ChatGPT Search

#910

Post by charcoalhobo »

dyauspitr wrote: post #708
GolfPopper wrote: post #602

Every time I see one of these topics, I go ask chat GPT a question to which I know the answer on a topic where I would like to be able to get useful answers to similar questions that I do not know the answer to. This time it was, "Did Paul Edwin Zimmer write a fourth Dark Border novel?" (Real answer: Yes, Ingulf the Mad . You can find the answer on his Wikipedia page.[1]) ChatGPT's[2] answer: "Yes, Paul Edwin Zimmer…

If you don’t trust the answer I always tell it to look it up and it invariably gets it right

Invariably seems like a strong word for the output of an LLM. If we can prompt a language model to double-check it's work (and that means it always gets it correct), why is that not part of the base prompt?
Post reply on HN