I use Kagi exclusively and refuse to offload my brain to a thing which has no accuracy guarantee ever. The emitted answers to the queries it has given can be completely bogus, and developers of these things low key expect me to believe what their black box say? Nah, never.
Instead I use a search engine and do my own reading and filtering. This way I learn what I'm researching, too, so I don't fall into the vicious cycle of drug abu ^H^H^H^H^H laziness. Otherwise I'll inevitably rely more on more on that thing, and be a prisoner of my own doing by increasingly offloading my tasks to a black box and be dependent on it.
Google recently (unrequested) provided me with very detailed AI generated instructions for server config - instructions that would have completely blown away the server. There will be someone out there who just follows the bouncing ball, I hope they've got good friends, understanding colleagues, and good backups!
Kagi is a search engine, which is developing its own index, and use the other indexes to give a more comprehendsive result if their own index doesn't fulfill the query you made.
Moreover, Kagi is a paid service. It has no ads, no hidden ranking, nothing to earn money by manipulating you. On the contrary you, the user, can add filters and ranking modifiers to promote the sites you find to be useful/truthful and demote others which push slop and SEO optimized content to your eyeballs. This is per user, and is not meddled with.
This makes Kagi very deterministic (unlike LLMs), very controllable (unlike LLMs), and very personalized (unlike LLMs). Moreover, Kagi gives you ~20 results or so per search, and no fillers (again, unlike LLMs).
I don't use Kagi's AI assistance features, and I don't pay for the "assistant" part of it, either.
I don't offload my brain to Kagi, because I don't prompt it until it gives me something I like. Instead, I get the results, read them, learn what I'm looking for, and possibly document what I got out from that research. This usage pattern, is again very different than prompting an LLM until it gives you something somewhat works or sounds plausible.
I do the hard work of synthesizing and understanding the answer. Not reading some slop and accepting it at face value.
> I don't offload my brain to Kagi, because I don't prompt it until it gives me something I like.
Similarly, I don't offload my brain to LLMs.
> I do the hard work of synthesizing and understanding the answer. Not reading some slop and accepting it at face value.
Again, it's not necessary to accept LLM output at face value.
Use tools, think for yourself, sure. This applies to various tools: Kagi, LLMs, and others. None of these give you "accuracy guarantees". You usually have to think for yourself.
My favourite example of a situation where you don't have to think for yourself is asking an LLM to implement a function in a very strongly typed language. There only is one implementation of `a -> a`. For `(a -> b) -> List a -> List b`, you could return an empty list instead of performing map. There aren't that many implementations of `(a -> b -> b) -> b -> List a -> b` (three as far as I can see: left/right fold and a function which just returns the accumulator). It's easier to verify the LLM solution than to implement it yourself!
Instead I use a search engine and do my own reading and filtering. This way I learn what I'm researching, too, so I don't fall into the vicious cycle of drug abu ^H^H^H^H^H laziness. Otherwise I'll inevitably rely more on more on that thing, and be a prisoner of my own doing by increasingly offloading my tasks to a black box and be dependent on it.