5 ms·

Fails my personal benchmark, unfortunately: haskell megaparsec operators -> should return [0] somewhere high in the result set. Found it on the third page of B

Fails my personal benchmark, unfortunately: haskell megaparsec operators -> should return [0] somewhere high in the result set. Found it on the third page of Brave Search. For comparison, it's the very first result on Google, second page for Bing, and the third page for DuckDuckGo. Sure, it's a bit obscure, and I don't even write that much Haskell these days. But it's calibrated to determine whether I can rely on the search engine to quickly surface what I need to be productive, and whether I can trust its 'negative results' (I don't see what I want on the results page -> I need to refine my query or take a different approach). The version number returned in the URL also shows how well the engine handles keeping up with versioned documentation pages and aggregating 'link juice' between them. I try this on every search engine alternative that pops up on HN. Very, very few pass. I find it often works like an adversarial example, yielding completely nonsensical results. [0] https://hackage.haskell.org/package/megaparsec-5.2.0/docs/Text-Megaparsec-Expr.html https://hackage.haskell.org/package/megaparsec-5.2.0/docs/Te... (or any other megaparsec-#.#.# version)

burkaman 3y ago
That's the first result on Kagi, but I think that's because they just call Google behind the scenes for you. Edit: I didn't mean this to be an insult, I pay for Kagi and use it exclusively. But actually, the Kagi and Google first page of results look fairly different for this query, so Kagi is doing more curation and ordering than I realized.
- Spivak 3y ago
  That's pretty much how all small search engines work, but with varying backing indexes. Kagi's value-add is Google quality index (since they literally pay for it), no ads, their ranking algorithm, and tools to control the results that Google doesn't give you like blacklists and your own weights. $25/mo is unfortunately a little steep because their lower priced option with 700 searches/month is comically small. On just my work browser I have 4000 Google searches last month. Doesn't count my personal laptop or phone. It's really hard to compete with $0/mo. for Google + AdBlock. I really do like Kagi better but $25/mo. is Tailscale + Notion + Spotify.
  - eitland 3y ago
    I have the early adopter professional plan, but I must admit that while I consider myself very much a power user I have never been close to even 700 searches a month.
    - mathgorges 3y ago
      Can you walk me through what a typical search day is like for you? To me, the 700 soft limit seems absurd so I'm curious to hear how you interact with the product. I average 50-100 searches per work day and 10-20 searcher per off day (totalling 1-2.5k per month). But I don't feel I'm using search in a particularly strange way. For example, if I need to know how the new WidgetBean in SpringBoot works I'll usually end up making ~10 queries related to it before I move on to a new area of research. E.g., I'll search "SpringBoot WidgetBean release notes" then "SpringBoot WidgetBean examples" or "WidgetBean test double", "WidgetBean in MockMvc test", "WidgetBean PowerMock known issues in MockMvc test", "WidgetBean FooService interation in MockMvc test SpringBoot 3". Essentially, when I'm searching I go wide to survey the information landscape, then refine my search term as I discover what I actually need to know This quickly balloons as I'm expected to know many things at $dayjob :) What does it look like when you search?
      - eitland 3y ago
        I'll try to get back with some numbers later today.
        eitland 3y ago
        This month, sorry for the crazy date formatting, I copied it from the Kagi UI: Date (UTC) Searches Apr 28, 2023 23 Apr 27, 2023 11 Apr 26, 2023 24 Apr 25, 2023 4 Apr 24, 2023 18 Apr 23, 2023 7 Apr 22, 2023 1 Apr 21, 2023 6 Apr 20, 2023 7 Apr 19, 2023 10 Apr 17, 2023 18 Apr 16, 2023 1 Apr 15, 2023 14 Apr 14, 2023 22 Apr 13, 2023 7 Apr 12, 2023 18 Apr 11, 2023 11 Apr 10, 2023 13 Apr 9, 2023 4 Apr 8, 2023 15 Apr 7, 2023 7 Apr 6, 2023 6 Apr 5, 2023 7 Apr 4, 2023 12 Apr 3, 2023 19 Apr 2, 2023 11 Apr 1, 2023 15 The last few months: (UTC) Searches Apr 2023 311 Mar 2023 495 Feb 2023 313 Jan 2023 434 Dec 2022 386 Nov 2022 396 Oct 2022 425 How I search: I typically start with one of the same type of queries as you did above, then I open every interesting-looking page from the first query in a nested tab (I use tree style tabs in my "research" browser) and start reading them, ctrl-clicking again to open interesting links in new tabs nested in a new level beneath, closing pages that was clearly useless and bookmarking in raindrop if I immediately sense that the oage I am on is something I will need again. Once I have exhausted this tree, at least temporarily, I start a new search - like you do - in a new top level tab anf create a new tree from it.
- eitland 3y ago
  They have API access to Google but there is clearly a sprinkling of magic and logic between. How can I know? The magic is proven by the fact that Kagi mostly respect my queries (or accept my bug reports if I can prove they didn't) despite being built on the shaky foundation of Bing and modern Google. Also there is some good old fashioned engineering there, like allowing me to pin, prioritize higher or lower or block certain sites.
seti0Cha 3y ago
I don't know that I would rely on performance on a single search to categorize how a search engine performs even in related areas to that one search. Also, maybe this is a stupid question, but are you letting Google personalize your results? If so, the comparison becomes even more problematic.
autoexec 3y ago
Brave search also failed the old "office -microsoft" search by returning a ton of microsoft pages. Just to be safe I tried "office NOT microsoft" too but it was even worse. DDG and Bing (no surprise there) fails at it too, but Google actually works.
George83728 3y ago
[dead]

burkaman 3y ago

That's the first result on Kagi, but I think that's because they just call Google behind the scenes for you. Edit: I didn't mean this to be an insult, I pay for Kagi and use it exclusively. But actually, the Kagi and Google first page of results look fairly different for this query, so Kagi is doing more curation and ordering than I realized.

Spivak 3y ago
That's pretty much how all small search engines work, but with varying backing indexes. Kagi's value-add is Google quality index (since they literally pay for it), no ads, their ranking algorithm, and tools to control the results that Google doesn't give you like blacklists and your own weights. $25/mo is unfortunately a little steep because their lower priced option with 700 searches/month is comically small. On just my work browser I have 4000 Google searches last month. Doesn't count my personal laptop or phone. It's really hard to compete with $0/mo. for Google + AdBlock. I really do like Kagi better but $25/mo. is Tailscale + Notion + Spotify.
- eitland 3y ago
  I have the early adopter professional plan, but I must admit that while I consider myself very much a power user I have never been close to even 700 searches a month.
  - mathgorges 3y ago
    Can you walk me through what a typical search day is like for you? To me, the 700 soft limit seems absurd so I'm curious to hear how you interact with the product. I average 50-100 searches per work day and 10-20 searcher per off day (totalling 1-2.5k per month). But I don't feel I'm using search in a particularly strange way. For example, if I need to know how the new WidgetBean in SpringBoot works I'll usually end up making ~10 queries related to it before I move on to a new area of research. E.g., I'll search "SpringBoot WidgetBean release notes" then "SpringBoot WidgetBean examples" or "WidgetBean test double", "WidgetBean in MockMvc test", "WidgetBean PowerMock known issues in MockMvc test", "WidgetBean FooService interation in MockMvc test SpringBoot 3". Essentially, when I'm searching I go wide to survey the information landscape, then refine my search term as I discover what I actually need to know This quickly balloons as I'm expected to know many things at $dayjob :) What does it look like when you search?
    - eitland 3y ago
      I'll try to get back with some numbers later today.
      - eitland 3y ago
        This month, sorry for the crazy date formatting, I copied it from the Kagi UI: Date (UTC) Searches Apr 28, 2023 23 Apr 27, 2023 11 Apr 26, 2023 24 Apr 25, 2023 4 Apr 24, 2023 18 Apr 23, 2023 7 Apr 22, 2023 1 Apr 21, 2023 6 Apr 20, 2023 7 Apr 19, 2023 10 Apr 17, 2023 18 Apr 16, 2023 1 Apr 15, 2023 14 Apr 14, 2023 22 Apr 13, 2023 7 Apr 12, 2023 18 Apr 11, 2023 11 Apr 10, 2023 13 Apr 9, 2023 4 Apr 8, 2023 15 Apr 7, 2023 7 Apr 6, 2023 6 Apr 5, 2023 7 Apr 4, 2023 12 Apr 3, 2023 19 Apr 2, 2023 11 Apr 1, 2023 15 The last few months: (UTC) Searches Apr 2023 311 Mar 2023 495 Feb 2023 313 Jan 2023 434 Dec 2022 386 Nov 2022 396 Oct 2022 425 How I search: I typically start with one of the same type of queries as you did above, then I open every interesting-looking page from the first query in a nested tab (I use tree style tabs in my "research" browser) and start reading them, ctrl-clicking again to open interesting links in new tabs nested in a new level beneath, closing pages that was clearly useless and bookmarking in raindrop if I immediately sense that the oage I am on is something I will need again. Once I have exhausted this tree, at least temporarily, I start a new search - like you do - in a new top level tab anf create a new tree from it.
eitland 3y ago
They have API access to Google but there is clearly a sprinkling of magic and logic between. How can I know? The magic is proven by the fact that Kagi mostly respect my queries (or accept my bug reports if I can prove they didn't) despite being built on the shaky foundation of Bing and modern Google. Also there is some good old fashioned engineering there, like allowing me to pin, prioritize higher or lower or block certain sites.

seti0Cha 3y ago

I don't know that I would rely on performance on a single search to categorize how a search engine performs even in related areas to that one search. Also, maybe this is a stupid question, but are you letting Google personalize your results? If so, the comparison becomes even more problematic.

autoexec 3y ago

Brave search also failed the old "office -microsoft" search by returning a ton of microsoft pages. Just to be safe I tried "office NOT microsoft" too but it was even worse. DDG and Bing (no surprise there) fails at it too, but Google actually works.

George83728 3y ago

[dead]