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

Privacy Pass Authentication for Kagi Search

blog.kagi.com

Post reply on HN
Print view View original

191–200 of 359 posts

Re: Privacy Pass Authentication for Kagi Search

#191

Post by g-b-r »

MostlyStable wrote: post #100

One of the biggest complaints about Kagi from people who have not yet adopted it is their privacy concerns around having to login and have payment information. I'm not one of the people that has been concerned about that, but I'm curious to what extent this alleviates those concerns among those that have had them.

> I'm not one of the people that has been concerned about that, but I'm curious to what extent this alleviates those concerns among those that have had them.

I am, it's mind-blowing to me that anyone would login to a search engine (yes, I know how many do it, now).

After a brief verification of the system, I'm pretty sure I'll sign up, now

Re: Privacy Pass Authentication for Kagi Search

#192

Post by tansan78 »

I am just thinking there might be other better ways to preserve user's search privacy: using LLM embeddings (https://en.wikipedia.org/wiki/Word_embedding).

The browser creates embeddings of user query, then send the embeddings to the server.

To complete a search, the server is a machine and it does not really need text to understand what a user want. A series of numbers, like LLM embeddings, are totally fine (actually it might even be better, because embeddings map similar words closely, like Duck and Bird have similar embeddings).

On the privacy side, LLM embeddings are a bunch of numbers. Even the embeddings are associated with a user, other people cannot make meaning out of the embeddings. Therefore the user's privacy is preserved.

What do you think?

Re: Privacy Pass Authentication for Kagi Search

#193

Post by faeranne »

MostlyStable wrote: post #100

One of the biggest complaints about Kagi from people who have not yet adopted it is their privacy concerns around having to login and have payment information. I'm not one of the people that has been concerned about that, but I'm curious to what extent this alleviates those concerns among those that have had them.

Assuming the cryptography does what they say it does (am not a cryptography expert, so I can't verify that part), this would completely disjoin a search request from any account info. The account generates several "search tokens", and for each search request, one of those tokens is spent. The tokens are generated on-device, and until spent, never leave the device, so in theory there's no way for Kagi to know which account generated the token just from the token alone. This doesn't fix fingerprinting or IP associations (though the plugin for Firefox and Chrome supposedly takes efforts to try and limit fingerprinting too), but this isn't any better/worse than simply using Google or Duckduckgo, and functions on Tor if you really want some privacy.

Again, not sure on how the tokens are proven legit without ever sharing them, but there's probably some ~~zero-knowledge proof~~ stuff going on that covers that.

Edit: Not zero-knowledge proof. Seems to be Blind Signature?

Re: Privacy Pass Authentication for Kagi Search

#194

Post by faeranne »

sebazzz wrote: post #40
ajayyy wrote: post #9

Earlier quoted context omitted.

The tokens are "generated" on the client, and the server just gives the client enough information to make that locally generated token become "valid", without being able to link that token to a specific validation attempt

So basically the server signs the token and afterwards the server can verify its own signature for every request with that token?

looking at it from a high level, it doesn't appear the final token ever leaves the client till it's being redeemed. There's a middle step that does get signed, but this part is not what is sent.

Re: Privacy Pass Authentication for Kagi Search

#195

Post by faeranne »

echoangle wrote: post #132
wasabi991011 wrote: post #117

Earlier quoted context omitted.

In the simplest terms, the token generation process B->C is done with the user's private key. So even if the server knows A,X,B they can't link it to the token C.

But if the server is allowed to vary X, it can basically act like different servers to each client, and can then when given a token check for which server would have been valid. The solution I got from the other replies is to make sure that the server uses the same X for everyone by verifying it as a client.

It appears that the public side of X is sent as the first part of the handshake, without any login info yet, and can be verified as part of B, thus a varying X would be easy to detect... I think.

Re: Privacy Pass Authentication for Kagi Search

#197

Post by kortilla »

laurex wrote: post #163
baridbelmedar wrote: post #157

Earlier quoted context omitted.

Playing devil's advocate... Yeah, the ad supported model has its problems, but it also makes the internet way more accessible. If we think about it, companies and people with more money are basically subsidizing these services for everyone else. They're the ones seeing the ads that keeps the lights on for users who can't afford to pay. If everything was subscription only, a ton of people like students, low income fam…

Playing the... angel's advocate... There's no reason why a subscription model could not also be used to subsidize people who can not pay, other than that companies are structured to extract as much as possible (by law, if they are public). There are good network effect arguments about why this strategy can be effective, not simply 'altruistic.' Ads simply make the extraction happen across the board, except that the a…

> structured to extract as much as possible (by law, if they are public).

This is not true and it’s not what fiduciary duty means. Stop repeating it, it’s really dumb.

Companies very frequently do not monetize things that they could under the guise of “building brand recognition” or “establishing a user base”. It’s even as easy as “raising the price will alienate customers we think are important to long term revenue”.

It’s trivial to justify not extracting maximum price and public companies do it all of the time.

Look at Costco’s business model if you want an example

Re: Privacy Pass Authentication for Kagi Search

#198

Post by jorvi »

laurex wrote: post #163
baridbelmedar wrote: post #157

Earlier quoted context omitted.

Playing devil's advocate... Yeah, the ad supported model has its problems, but it also makes the internet way more accessible. If we think about it, companies and people with more money are basically subsidizing these services for everyone else. They're the ones seeing the ads that keeps the lights on for users who can't afford to pay. If everything was subscription only, a ton of people like students, low income fam…

Playing the... angel's advocate... There's no reason why a subscription model could not also be used to subsidize people who can not pay, other than that companies are structured to extract as much as possible (by law, if they are public). There are good network effect arguments about why this strategy can be effective, not simply 'altruistic.' Ads simply make the extraction happen across the board, except that the a…

I mean, we could also just direct-pay websites (for example with Brave's Basic Attention Token model).

Imagine a utopian world where you just pay per site visit, and in return all companies selling stuff don't have an inflated advertising budget and free market effects force them to pass the savings on to you, meaning the net cost increase for you is zero. And as a side-effect, quality products float to the top, since you hear of them mostly by word-of-mouth, meaning products compete on value-per-dollar.

Sadly human psychology and economics does not work that way haha. We pay what the market will bear, and increasing sales via a torrent of ads is cheaper than increasing the value-per-dollar ratio of the product.

Re: Privacy Pass Authentication for Kagi Search

#199

Post by 3s »

tansan78 wrote: post #192

I am just thinking there might be other better ways to preserve user's search privacy: using LLM embeddings ( https://en.wikipedia.org/wiki/Word_embedding ). The browser creates embeddings of user query, then send the embeddings to the server. To complete a search, the server is a machine and it does not really need text to understand what a user want. A series of numbers, like LLM embeddings, are totally fine (actua…

This is an interesting idea, and indeed such an approach to providing privacy has been formalized to different degrees and varying levels of success (eg. [1][2]).

[1] https://arxiv.org/abs/1204.2136 [2] https://arxiv.org/abs/2210.03458

Unfortunately, as described, such a solution would only satisfy a somewhat meaningless notion of privacy. Specifically, the embeddings by definition contain potentially private information about the user, revealing things like "I'm asking about birds" to use your example. Even though it might "compress" the query in a slightly lossy way, it would still reveal a great deal of information about the query.

A true solution to this problem would require something like differential privacy and adding noise to the embeddings. However, the noise required would (likely) end up destroying too much information from the embedding to preserve accuracy of the LLM.

Re: Privacy Pass Authentication for Kagi Search

#200

Post by WD-42 »

freediver wrote: post #188
drdaeman wrote: post #28

Neat! It's rare to see that a service you use actually does something that benefits the user rather that itself. An unexpected, but a really pleasant surprise. I wish this extension would integrate better with the browser by automatically understanding the context. That is, if I'm in a "regular" mode it'll use my session, but if I'm in a "private browsing" mode (`browser.extension.inIncognitoContext`) it'll use Priva…

> (I don't use Orion, as there's no GNU/Linux version.) We commenced work on Orion for Linux yesterday.

Amazing!!!
Post reply on HN