I didn't use AI because for these things because for some things (database design, constructing a grammar) you _really_ want to understand your code completely, and I don't want to deal with hallucinations.
Yes, GitHub code search is much better than Googling for the same information.
And ChatGPT can't show you the whole codebase which is important to understand structure - plus sometimes it's easier to surf around a few files than have ChatGPT print at you.
Entirely off topic, but I hope referring to AI art as "generative art" doesn't catch on
There's such a great and rich world of generative art (explorations of algorithmic aesthetics, not AI generated) which is probably already endangered by the influx of the recent AI boon - would be a shame for it to be further buried by naming overload
That's a really interesting point, "content creators" always gave me a weird feeling but I never really thought on it, assuming it was just because it reeked of euphemism
Odd that we've collectively ended up giving human creativity a term that sounds like a tool or automaton, while giving the automatons our term.
I tried Kagi and even signed up for the basic plan but ultimately cancelled. If the basic plan included twice as many searches per month, I'd be all in. However, I find Startpage results are "just as good" and free. I didn't use a lot of the other Kagi tools.
So make it cost less or offer more and I'm in Kagi!
I've been using kagi and that's my biggest complaint. Upgrading from the basic plan seems like a lot of money but being on the basic plan, I have mental friction every time I do a search because I have anxiety about running out of searches. I actually haven't ran out in the 3 months I've been using it for but I don't like the anxiety 300 searches/month creates for me and maybe I would run out if I didn't have the anxiety limiting me
I just paid for the new unlimited plan, and haven’t looked back. I agree about the 300 searches per month thing. I feel like I run many more than that, but haven’t counted.
I just went back to starter from ultimate because they are now adding 19%(?)VAT in Germany and so I resumed paying for gpt4 which adds dall e for 7 euro less (22 openai, 29 Kagi ultimate) IF I stick to 10 searches a day. I forgot about this new distribution setup immediately and prompted by this thread I now find that in the first 6 days back on starter I used 11 searches per day so there is hope.
I had tried using Kagi just for home use since the bulk of my searches are work related (Kagi is not something I can claim as an expense, before you ask) but still found little added value beyond "I think I like this more".
I did also try a bunch of side by side searches "for work" (so mostly leading to stack overflow) and they were the same results, more or less, for my searches between Kagi and Startpage.
One thing I found myself doing as I became conscious of my search quota limitations is going directly to sites and searching there using the local search. This doesn't seem like a bad thing but it's one more step in an already friction filled process. :)
For me LLMs feel like a lecture where you understand all the words and ideas but once you leave you realise you really haven't understood anything new but you perhaps have some new avenues to explore. More of an advanced exploratory tool.
One tip for Google Search is to add `tbs=li:1` as a query parameter to your search shortcut to activate verbatim search. You can also use `num=100` to get more results at once.
No, it does the same thing as selecting Tools > [All results >] Verbatim.
Putting quotes in addition is still necessary when searching for a phrase, or when you have multiple search terms you want all to appear in every hit (though that's still not 100% reliable).
Keyword search without SEO trash is EXACTLY what I want, and Marginalia sounds like it might fit the bill. Looking forward to seeing how well it performs over the next couple of weeks!
Definitely not to detract from using Marginalia directly, but I think it's worth noting that Kagi actually uses them as one of several external sources[0]. I think the Small Web filter leans on their results.
It's still a bit rough around the edges and technically limited. Index is pretty small too, definitely no straight up replacement for anything, but surprisingly often it provides a useful minority report where other engines fail.
I only search if GPT4 does not provide what I need, which in about 80% of cases it does. My use cases tend to be ones where it’s fairly obvious if the results are true or not, which, once again, they are in the vast majority of my requests, and even when GPT4 is wrong it’s often “usefully wrong” in the sense that it gives me better idea of what to search for in traditional search engines.
Interesting to see Kagi on this list. One of our users for Zenfetch specifically requested the option to see their Zenfetch articles alongside their search results, so we naively developed that feature to appear beside Google SERPs...
Turns out, he was a Kagi power user. Not the worst mistake on our end, though pretty neat to see it in the wild
I've been using Brave Search[0], and I have to say I've been quite impressed. I don't know if it will be enshitified going forward, but for now they run their own index and have been doing quite well for my purposes.
I'd use Kagi, but I really dislike the idea of my searches being linked to my credit card and home address for privacy reasons. It's a shame, because I do believe that paying a reasonable price for services is often more sustainable and, paradoxically, consumer-friendly than the alternative.
While I've a tonne of good will towards Kagi and its mission and appreciate that you've enshrined this in your privacy policy, it's sufficient that a service _can_ link my searches to my physical person for me not to use it.
That said, I acknowledge this is, for many, an unreasonably high bar.
I recently did this to figure out how to insert JSON into Sqlite's FTS5 https://github.com/brouberol/5esheets/pull/292
I didn't use AI because for these things because for some things (database design, constructing a grammar) you _really_ want to understand your code completely, and I don't want to deal with hallucinations.