Hacker News Reader

Comment by rand_flip_bit | Hacker News Reader

Ughh, please don’t. The last thing I want in my search engine is AI, especially for a “privacy focused” one like Kagi. Paying to have all my search queries get routed through ChatGPT (and effectively used as training data) is absolutely not what I want. I suspect it’s too late though, Kagi has already drank the AI koolaid.

View on HN

Your search queries on Kagi are paired with your credit card, so if that doesn't bother you, ChatGTP shouldn't either.

I'm wondering why this doesn't seem to bother anyone. I always use a browser profile that's not signed into anything when searching, whether that's for work or personal. I know this does not guarantee privacy, but why make it easy for them?

The whole point of using Kagi for me is that I've got my search options customised and I'm paying for a service better than the alternatives. The payment privacy aspect of that lives on a completely different layer: you'd need to first solve the KYC -vs- privacy problem, and that's more of a politics problem than tech.

I don't care what kind of problem it is. That doesn't change the threat model.

I understand and appreciate what Kagi trying to do. But linking search queries with your real name is huge privacy invasion. Even though Vlad says they do not log queries. Well... Without proof there could be only trust. Decision is up to you.

For sensitive searches I do recommend only Tor. For casual searching SearX. Anonymous or not in regular browser doesn't change much if you don't use any kind of IP hiding.

Koolaid or not, as long as they're sticking to Claude (or at least away from MS/Google/Meta) I don't really mind. FastGPT[1], which recently became unaccidentally public[2], seems to work pretty well, even though it's still separate from their main offering. The feature creep of their main search engine, like the podcast search and listicle stuff, is easy to toggle off, so I imagine LLM integration will be optional as well.

[1] https://news.ycombinator.com/item?id=35719482

[2] https://labs.kagi.com/fastgpt

I mean yeah, credit where credit is due Kagi lets you turn all these things off and customize how results are produced and displayed like no other.

> and effectively used as training data

Models are not trained during inference, are they?

Regardless, it seems like LLMs are going to have some role to play in search in the future. It's just not clear what that role is yet.

100% the data you input to the ChatGPT 4 prompt window is being used to train ChatGPT 5, there is a reason a lot of companies are placing bans on it and telling employees to never enter sensitive data.

I sincerely hope that LLMs won’t play a role in search, I don’t need a search engine trying to infer what I “meant” to type, Google and DDG started doing that and pushed me to Kagi as a result. Just take the words I type verbatim and search them in your index, don’t assume this is my first time using the internet.

only in the chat app. Per TOS, data sent to APIs is not used for training. All thesee services undoubtedly use the API

Reply on news.ycombinator.com