Hacker News Reader

Comment by throwup238 | Hacker News Reader

I wonder if Kagi's implementation of the same feature influenced anyone at google or if this was an inevitable development. Google has been providing the AI answer boxes for years but it's never been very user driven.

With Kagi you can use !fast/!expert bangs to use one of several LLMs fed the top few search results or as of last month, just end the search query with a question mark (no affiliation just a happy customer). It's almost completely changed how I use search.

View on HN

I pay for Kagi, I'm only vaguely aware that they have AI features and I have no interest in using them. What I like is that it doesn't annoy me about that stuff or anything, it just lets me use it (kagi search) as I want. Completely different from aggressive big tech "you're the product" companies that use modal "got it?" popups to try and push features you don't want. I hope that doesn't change for Kagi more than anything else.

Also, I rarely use search to answer questions and anyway would never just trust an answer. I even read SO before pasting in the code and decide if it fits with what I'm doing and what modifications the answer may need, I guess I'm old. But more importantly, I use "search" mostly to get to pages I know exist or expect to exist. And getting to the right page faster is what I care about, not any answer.

I pay for Kagi and was similarly only vaguely aware that they have AI features and I had no interest in using them.

I started adding a question mark to my query out of curiosity and their instant answers are really good. Also they have links to their references, which I often check to verify the answer.

You should try Kagi's FastGPT [0]. You need not trust the answer, as it gives the URL sources for each paragraph to verify yourself.

[0]: kagi.com/fastgpt

To me, this is the key UX component that Kagi does correctly -- traceability to the underlying source result (which is usually also quite good with its linkbait penalizing).

I've found Kagi to be a great start for "I want to know a current fact in an area I don't follow actively."

It's quite interesting what a search engine can be in 2024 when the product isn't the one using it, but the search results themselves.

Kagi with a wayy smaller budget than Google have really managed to make something pretty cool. Still does not replace google for 100% of my searches, but for the ones it does it is remarkably good, and so much less mentally taxing when working on a coding problem. I don't have to on top of everything manually filter search results, I can just click the top link.

Kagi uses the Google search API.

"Our data includes anonymized API calls to traditional search indexes like Google, Yandex, Mojeek and Brave, specialized search engines like Marginalia, and sources of vertical information like Wolfram Alpha, Apple, Wikipedia, Open Meteo, Yelp, TripAdvisor and other APIs. Typically every search query on Kagi will call a number of different sources at the same time, all with the purpose of bringing the best possible search results to the user."

So that's only telling a small part of the story.

Not to mention that they have their own index that they're constantly working to expand:

> But most importantly, we are known for our unique results, coming from our web index (internal name - Teclis) and news index (internal name - TinyGem). Kagi's indexes provide unique results that help you discover non-commercial websites and "small web" discussions surrounding a particular topic. Kagi's Teclis and TinyGem indexes are both available as an API.

https://help.kagi.com/kagi/search-details/search-sources.htm...

The comment I replied to implied that they have their own large index equivalent to Google's and so don't need to use Google. It's not true, and I bet you would find that a very significant portion of the search results come directly from Google queries behind the scenes.

I don’t think that implication was in the comment you replied to. “Well they just use google anyway” is too reductive.

This brings an interesting dilemma. Google could try the Kagi approach.

If you're sitting in the CEO chair at Google, you're looking at a tough decision. Google Search isn't just another product; it's the heart of the business, raking in 60% of the revenue. They've been king of the search hill for over 20 years because they did one thing better than anyone else: search. Now, with LLMs entering the scene, adding them to search results could slash ad views and revenue. Remember the days when a simple search would send you down a fascinating rabbit hole of articles? Those days could be numbered.

But here's the kicker: if you resist integrating LLMs into Google Search, you might slow down the revenue decline, sure. However, sticking to the old ways might also get you branded a Luddite and cost you your job. It's a classic tech dilemma - innovate or perish.

Google has been on a mission for years to stop letting users click through to other sites.

They started with the info boxes that extracted info from sites so you don’t have to click and read.

This is just a natural evolution of that.

People don't want to click through, they want their question answered.

Does Google pass through ad revenue to its sources for searches serviced by instant answers?

It seems like a simple business model: you give me answers, I give you a cut of the money I make off them, and I keep a piece of that money.

I think there are different classes of result that we should discuss. First, there are pure facts. I know that weather.com wants to interpose between me and the weather forecast that my government produced for me using my tax money, but Google doesn't need to assent to weather.com's scheme. That is something they can just answer. In the same way, whether my local chaat stand is open on Tuesdays in not a fact that belongs to Yelp.

Then there are materials that are freely available but someone wants to stand in the way of them. For example, all of the wikipedia-but-with-ads hosts out there. Do we owe them anything? I say no. Do we owe anything to all of the journal indexes that try to convince you to pay $26 for the PDF of a paper from 1913 that's in the public domain anyway? I think not. And I don't think it is problematic if I ask Google whether Heidegger was a nazi, and Google answers with an answer instead of referring me to the gatekeepers of centrury-old information.

There's a third class too, perhaps the most important one -- facts which take time, money, and/or skill to assemble.

E.g. a summary of the minutes from your local city council, a respin of government economic statistics that answers a topical question, or a news investigation into a topic.

I'm ambivalent as to the business model that sustains that third class, but I resolutely think one needs to exist.

And Google, as currently constructed and operated, is not it. It doesn't originate content.

Ergo, passthrough of Google revenue to some entity that can / does do that seems reasonable.

> a summary of the minutes from your local city council, a respin of government economic statistics that answers a topical question, or a news investigation into a topic.

The city council minutes would likely be published by the city. Nobody is entitled to revenue from that, it's paid for with tax dollars.

Government economics statistics are also published openly. I happened to have asked a bunch of questions about life expectancy and GDP recently and Google answered those questions with info boxes and links to the data sources. If I ask a question which can be answered in a few words, I'm perfectly content not to have to view ads to read an article from an economics wonk to manually find the tiny nugget I'm after.

And as for a news investigation, there's almost no question you could ask in isolation that would be better answered by reading a full article, I think. The person asking "Who is facing legal action in the aftermath of the Sandy Hook massacre?" wants to get back "Alex Jones." They almost certainly didn't want to read a full article, and if they did, they'd click on it.

Which is to say, Google is doing nontrivial work to accomplish a specific outcome that the "third class" sources don't solve (and usually don't even try to solve). Besides, how do you know the content Google is pulling from...

1. Came from a single source and was not corroborated with multiple sources?

2. Wasn't just scraped from somewhere else or compiled with AI?

3. Isn't wrong? Should sources get paid for incorrect information?

A passthrough system creates perverse financial incentives to answer questions that someone might someday have, without caring about whether the answers to those questions are genuine and ethically produced. And frankly, that's far worse than what we have now.

> The city council minutes would likely be published by the city.

Have you tried to look for yours, in a smaller city?

> Government economics statistics are also published openly

Certain statistics are published openly. However, the most useful derived statistics are a result of combining these with other data sets.

> And as for a news investigation, there's almost no question you could ask in isolation that would be better answered by reading a full article, I think

I'm not arguing excerpt vs full article. I'm arguing as to whether or not the full article exists in the first place.

There is a huge amount of critical, socially-useful information that requires effort, money, or skill to generate.

Previously, journalists created it.

Now, nobody does.

> Which is to say, Google is doing nontrivial work to accomplish a specific outcome...

What would Google be able to provide without content provided by underlying sources?

Nothing.

Google doesn't have reporters or posters. They provide platforms, then take a cut (or all) of the revenue generated by content on those platforms.

That's worth something, but it certainly doesn't deserve the 100% they're getting now.

> Besides, how do you know the content Google is pulling from...

Google knows. And if it doesn't, then it's unattributable content, which should open them up to legal liability.

ML/LLMs shouldn't be GPU-powered copyright washing machines: feed copyrighted content in, get uncopyrighted results out.

> A passthrough system creates perverse financial incentives to answer questions that someone might someday have, without caring about whether the answers to those questions are genuine and ethically produced. And frankly, that's far worse than what we have now.

You've literally described Google's current business model, and what they've turned the web into.

That's a slippery slope.

Should a professional programmer have to pay wikipedia if they read about an algorithm on it and then implement it and get paid for it?

It's not a slippery slope at all. Wikipedia has a very clear and explicit licensing model - content can be used under Creative Commons.

So do the sources that Google aggregates.

If I search for how old Taylor Swift is, is a clickbait website entitled to the revenue from my visit? If I'm searching for an objective fact, nobody owns that knowledge. Arguably if I want an objective answer to a question and Google gives it to me, they're the ones who did the work in this case.

Does ChatGPT which is what the op was comparing to?

I've been using both ChatGPT and Gemini (both the paid versions), more and more. My habit is to go search on Google, get frustrated I can't get a relevant result. As Gemini or ChatGPT and 2 of 3 times, get a useful answer.

At some point my habit will likely change to ChatGPT/Gemini first.

Like the OP. I type something cryptic (like keywords) into Google search vs typing 1 to 3 sentences into Gemini/ChatGPT. I should try typing the entire thing into Google search.

Interestingly this reminds me of stack overflow. Their search is legendarily bad but I'd often go start a new question, add tags, start typing my question in detail, and, in it's recommendations based on the content of my long form question it would find the relevant existing answer.

No, they don't. It's just a top position on Google. It's like showing up in the first position, but with way less click-through.

People want their question answered accurately, which is many times more likely to happen after clicking through than by relying on some opaque Google process that may or may not have interpreted the information correctly.

Doesn't that imply that they don't want a search-engine?

Can also hit ‘q’ to show the Quick Answer feature if you forgot the question mark.

Love this feature too, especially the references! I end almost all my queries with a question mark now.

Reply on news.ycombinator.com