I do the Kagi Ultimate thing which proxies over to ChatGPT and Claude and the like.
Kagi has always said it was privacy-focused, so I am actually curious if there's privacy associated with the AI chatbots as well? I don't think it can be completely private because if nothing else they're paying at a per-request level right?
Yeah I know, but presumably if I were doing something insane, like 1000 calls a minute, they'd have to do something to audit against? Even if they don't know what the content of the messages were specifically, they'd still see the tombert made 20,000 calls to GPT-4o.
I bet they just count them and show you a nice throttling page.
I wouldn't consider a simple counter against an account an invasion of privacy. And I'm pretty serious on privacy (I even set up 24/7 ollama server so I can avoid sending my more private LLM stuff to the cloud).
I really don’t like this framing at all, it’s misleading and dangerous.
This is basically just a VPN, and like a VPN, you hiding your IP isn’t a solution for “privacy”, it’s merely removing your IP as one identification vector.
The content of your interactions will still identify you, this has been established again and again.
Example:
In 2006, the Internet company AOL released a large excerpt from its web search query logs to the public. AOL did not identify users in the report, but personally identifiable information was present in many of the queries. This allowed some users to be identified by their search queries. Although AOL took down the file within a few days, it had already been widely copied and still remains available.
I think "anonymous" is as fair a framing as it gets. If you publish a novel anonymously, but reveal personal details about yourself in it or use a particular writing style that lets people identify you, that's arguably an expected outcome as well.
Just call it a proxy though, as that’s all it is. And end of the day, using better models with a VPN gives the same privacy and better performance. This is pointless beyond attempting to stop the bleeding from imploding search traffic.
DDG has made very poor technical and business decisions and they are trying to hold on but can’t escape the fact they are just a Bing wrapper. And Bing, like Google, is increasingly an index of AI spam.
Web search - from any company - is rapidly approaching full-on utility collapse.
Pushing half-truths about second and third tier LLM models being “private” will not stop the bleeding.
Which better models can you use with a VPN though, without signing up for an account?
Requiring a login makes all of your queries cross-correlate-able, unless you go through the significant pain of signing up for a new account very often (and even then, some LLM providers require SMS authentication for new accounts).
Given that, the fact that you can use DuckDuckGo without signing up from behind a VPN is pretty cool in my view.
Well you're missing about 4/5s of what they do to make it private so whenever you decide to join the rest of us with the whole picture we can talk about it.
Kagi has always said it was privacy-focused, so I am actually curious if there's privacy associated with the AI chatbots as well? I don't think it can be completely private because if nothing else they're paying at a per-request level right?