AI means the end of internet search as we've known it | Hacker News Reader

AI means the end of internet search as we've known it

technologyreview.com

40 pointsgnabgib58 comments

I was just trying to solve a configuration problem in Xcode today. I started with a web search, and found lots of proposes solutions, but nothing that worked for me. I asked ChatGPT, which regurgitated the same ideas I found on the web, plus a few more that also didn't work.

Finally I tried something undocumented on a hunch, and it kind of worked, and I shared my progress with a (human) colleague, who had the insight to take what I had done and finished a real solution.

Anecdotally, in mice, etc., etc.

Not to belabor a service frequently suggested here, but have you ever tried https://kagi.com/? I find it much more reliable.

This would be an opportune moment to enumerate the problem you had, the solutions in brief that didn’t work, and the one that did.

Some details, even just a link to the chat would be helpful. At the very least, you might help someone experiencing the same issue.

No, it doesn't.

AI has a lot of edge cases and caveats. It can be trivial like not being about to count the Rs in "strawberry". Or it can be more nefarious where it simply makes stuff up (eg some fake precedents in legal opinions). AI is still incapable of explaining its reasoning and dealing with errors.

Yes I know some of these problems like the "Rs in strawberry" problem have been solved but (IMHO) you're going to be dealing with those edge cases forever.

Another issue is response time. Currently, you need to go through several steps: query -> embedding -> LLM -> answer -> back to English. Each of these steps takes time.

But here's the big one: energy. The sheer scale of Google search needs to be put in context of how much energy is consumed and how many queries can be answered per unit energy. With all the steps involved in AI queries, we need orders of magnitude of improvement to compete.

Most searches are fairly simple. They just don't need a large model to answer them. There will absolutely be a place for AI queries and they will continue to get better but displace search? We're not even remotely close to that outcome.

AI responses need a source (lol, I know) or they just can't be trusted. It's that simple. Unless you believe everything you read on the Internet!

They obviously are a step of evolution beyond search in capability.

> The way AI can put together a well-reasoned answer to just about any kind of question, drawing on real-time data from across the web, just offers a better experience.

No? This experience is only better if the result isn't hallucinated nonsense, which the article acknowledged before, but then just ignored in the overconfident claim that nonsense is the future

If you're not willing to ignore reality and saturate your life with confabulated nonsense, then I don't think you're ready for the future.

Exactly. Replacing search with chatbots removes any opportunity to apply the media literacy I’ve spent decades learning. It gives every source the same sheen of correctness, making all information it gives essentially worthless.

It only does this if you don’t ask it to supply sources and double check where you aren’t confident. It allows you to cut through 99% of the bullshit within search results and double check where necessary. Perhaps it is a new type of media literacy but I don’t think it is too far off.

The idea that I might have to research the validity of a search result is very offputting. There was a time when I trusted Google to give me the most relevant result, filtering out the linkfarms and spam results. I don't see how AI gets us back to that trusting state.

Were you not already checking the validity of search results? Because Google's top few results I don't think were ever immune to "hallucination" where the top results happen to be garbage. It's where "don't trust everything you read on the internet" came from.

So I think the only thing that really needs to happen is just blindly trust the AI like you were apparently doing with early Google. I suspect however that you were gut checking Google which you can still do with any AI search that cites its sources.

You really are missing the experience of what google offered for their first decade, when pagerank seemed revolutionary in terms of filtering out spam and 101-level SEO tricks. We had working search engines for a while, and they were taken from us by, first, the providers in seeking greater ad revenue, and second, the AI floggers who want to exploit our now lowered expectations.

All you're doing here is justifying your lowered expectations.

My experience with Google is that the results were filled with SEO crap requiring quite a bit of manual sifting and confirmation/research (try to searching for reviews unbiased reviews that aren’t simply trying to get you to buy from Amazon using their affiliate link, or look for some unbiased nutrition information). ChatGPT will at least (with the right prompt) give me links to Pubmed or similar, rather than “Muscle Joe’s Nutrition House”.

The result only needs to be appealing to consumers. If hallucinated nonsense accomplishes that then job done!

Why is this written like the MIT reader hears about the technology for the first time? On other websites this would be more appealing.

Edit: Are we reading a generated article?

The quiet unsaid trick is that they expect you to pay 5x for the service because you need to feed the result back into other multiple models to double and triple check. Another tech year, another growth scam.

Ad blockers are going to need to become really advanced once chat bots start outputting sponsored answers or other injecting product recommendations into their usual output.

Oh you know the solution. Ad Blocker with AI

_Hey GPT, rewrite the following text removing all references to products or any ads. Keep the rest of the text verbatim._

The same product recommendation text could have come from a sincere human source, a biased human source, a hallucination, or a sponsorship. An ad blocker won't be able to tell the difference.

Wouldn't this be against a lot of consumer protection/advertising laws?

It's not like Google marks sponsored search results because they want to. They do it because they legally have to.

And if the LLM agent conforms to the law and clearly marks the output that is sponsored content then it should be trivial for ad blockers to filter it.

laws are for the little people, not big tech

AI doesn't connect you to communities like search engines do. Ask AI a question and it gives you an answer, no interaction with other people. Ask a search engine and it may bring up a forum with someone asking a similar question. Then maybe you join the discussion.

AI will bring about a lonelier online world.

That’s another issue they don’t want to talk about in their quest for money. They want to create a read-only single-page curated proxy in front of the web and sell it as an improvement.

I wrote about this a while back (https://matrix.ai/learn/blog/content-commoditization-and-tru...) arguing that SEO will transition to LLMO eventually. We won't bother optimising for search engine rankings, but instead for answers on LLMs.

Will that make search results useful again?

does that mean we can transition back to the useful search of 20 years ago!?

gazumped!

> won't bother optimising for search engine rankings, but instead for answers on LLMs.

the incentive for SEO is to drive click thrus.

What, if any, are the incentives for sites to optimize LLMs' answer? Would that not make click thru rates go down instead? I would actually imagine that the site would try to make their content anti-LLM, such that an LLM cannot sufficiently cover the content in their summary, and the user must end up visiting the site itself to verify.

The opposite, I imagine: if before, SEO was gaming page rank, then the new SEO will be gaming the LLMs to promote your site as a best result. It's still SEO'ed crap for clickthrus, it's just a different mode of noise diluting the signal.

Supply has to meet demand. I want my product featured as the answer to people's problems.

And yea "content" sites are probably going to want to be on their closed platforms. I was more talking about "solution" or "product" sites.

It's ironic considering how dependent LLMs are for search engines.

I doubt they're "ending", rather they will need to be re-born for RAG purposes.

I wrote about something tangential a few months ago; the need to have curation-based alternate forms of discovery on the WWW.

https://nadh.in/blog/decentralised-open-indexes/

Google Lens has already revolutionized my search. Anything that you can see but is difficult to describe, I now have a tool for searching for.

I use it a lot to track down original sources of videos and photos on Reddit, to check authenticity. I also use it to search for hardware components other manufacturers use, that I'd like to buy. Things like specific switches, knobs, faders, displays etc. (I build audio gadgets). I just highlight the part on the picture and then restrict my search to Alibaba, digikey or mouser. It's GREAT.

Lens is highly underrated.

I actually enjoy it. I still search with DDG, but Claude gives me direct answers without sifting through stackoverflow, or links upon links on how to do things on websites for their website functionality like e-bay or how to use things.

  Despite fewer clicks, copyright fights, and sometimes iffy answers, AI could unlock new ways to summon all the world’s knowledge.

Maybe a better title would be AI menas the end of knowledge as we've known it

I treat Copilot as glorified Bing search and nothing more. I try very hard not to pattern-match and let an output trigger my built-in human empathy and just have a question I want to answer, following the citations. I don't want to end up anthropomorphizing a lawnmower sometime down the line, even if the lawnmower can interject "please", "thank you" and "let me know if you need anything else" enough times to sound convincing.

I'm not as familiar with ollama so I don't know if the same approach would work.

I still prefer using search engines, but if I don't get an actual answer, I use LLMs with internet search. It is sometimes helpful for me, though most of the time, I get similar results as a normal search engine.

Some AI search companies [0] are even planning to add ads to their results, possibly on their free plans, which could make it harder for ad blockers to filter them out.

[0] https://techcrunch.com/2024/11/12/perplexity-brings-ads-to-i...

On one hand, I hardly touch a search engine any more. On the other, it's still a fairly common occurrence that I can give someone out in the world their ChatGPT experience.

I'm super curious to see how this ("the end of internet search as we know it") plays out with non-techy users. I strongly suspect that Google is "search" to many non-techy users the way that Internet Explorer was "the internet" to that same group.

Which I suppose leads to an obvious answer: device defaults will absolutely dictate what ends up being the go-to.

I think the sad thing is that you just don’t need search engines any more because there is too much conglomeration. There’s basically like three kinds of websites. Companies, Social Media, and Wikipedia.

It's just a matter of habit. It's all going to change once it is free and easily available to non-tech users. I wouldn't be surprised if using voice for communication with these bots started to be finally used. You just going to conversate with your own private search curator.

I am at the point I just browse arXiv because everything else sounds like bullshit.

Not that arXiv doesn't contain bullshit also but outside of that, in English at least, the internet has completely failed.

I have never used an LLM for anything. I use search engines dozens of times a day.

Complain about Google all you want, I prefer its link-supported AI summaries over Perplexity's mystery meat hallucinations with no links to human-written content.

Really? As shitty as Google can be these days, I and many others should prefer using a heavily redacted, controlled, essentially censored AI filter that gives me its results as a summary of what it got from the content that search engines often use anyhow, but with the possibility for random hallucinations right in the summary, and simply refusing to answer certain questions on what some corporate PR drones deemed "controversial" subject matter?

No thanks. Google may suck but at least it and similar platforms can lead me to random interesting links like Reddit and forum threads where actual humans give their human input on X or Y regardless of its nature.

Don't forget that narratives or advertising can also be added to that opaque output without disclosure. And that what can be exploited for profit inevitably will be.

I just asked ChatGPT where I can find an open restaurant near me, it told me to use Google Maps.

LLMs seem to have a pretty fundamental problem: they can't learn beyond their training data.

Also, can you really say Google isn't AI? Pretty sure there's a whole expert system lurking in there...

i swear i got it to do it once for the nearest mcdonalds,and it knew my location i didnt even have to tell it. havent been able to get it to do it since.

AI more likely begins the age of preemptive personalized search . . . a search tailored to people that happens before they even ask for it. Because that's best for consumers. And advertisers.

Alta Vista didn't do anything at all like that.

> AI more likely begins the age of preemptive personalized search

Maybe in a parallel universe. Right now, it continues the age of crappy search, with shitty search.

Yikes, this writer has made an article that harkens back to 2022.

The only consequence of AI that I made a prediction on was that the value of editing would increase. That people will stop accepting rambling articles and publications will stop publishing spam, whether that be complete nonsense using academic words or rearranging standard but uninteresting material.

Article like this keep proving me wrong.

Any theory that assumes humans are rational actors is doomed to fail.

On that day, it pushed me a story about a new drone company from Eric Schmidt. I recognized the story. Forbes had reported it exclusively, earlier in the week, but it had been locked behind a paywall. The image on Perplexity’s story looked identical to one from Forbes. The language and structure were quite similar. It was effectively the same story, but freely available to anyone on the internet. I texted a friend who had edited the original story to ask if Forbes had a deal with the startup to republish its content. But there was no deal. He was shocked and furious and, well, perplexed. He wasn’t alone. Forbes, the New York Times, and Condé Nast have now all sent the company cease-and-desist orders. News Corp is suing for damages.

Welcome to the next version of Google's mission to index all the world's knowledge and make money by serving ads against it, I guess, in this brave new world there doesn't need to even be the tiniest chance of a single cent getting to the person who wrote up the story.

IANAL, but recently I've become very interested in the possibility of a legal mechanism a bit like what the GPL does for code [0], a kind of poison-pill so if anybody uses your content to train their LLM, you can slap back by making everything that LLM ever emits into almost public-domain [1]. In other words, you strike 'em where it hurts, in the monetization.

[0] https://news.ycombinator.com/item?id=42582615

[1] https://en.wikipedia.org/wiki/Public-domain-equivalent_licen...

I think you missed the story about OpenAI wistleblower. /s

The AI response that Google puts at the top of search results now sounds confident but is sometimes just flat out wrong. I can’t trust it at all.

Reply on news.ycombinator.com