"Going back to the debate between folks like Xe, who believe that straightforward search queries are inundated with crap, and our thought leader, who believes that "the rending of garments about how even google search is terrible now is pretty overblown", it appears that Xe is correct."
Also, the article tested Mwmbl as well, not mentioned in the title here.
Search was the biggest feature of the web in the early '00s. Now it's such a mess. I can't imagine Search will ever be amazing again, given all the complexity of providing quality while still avoiding all the crap.
Yeah, the problem is that there is so much low quality content, that search doesn’t (or can’t) do a good job of surfacing it above the noise. There is still some signal left, but it’s such a small fraction that it’s much more difficult to filter it out.
Having said that, I’m usually still able to find what I’m looking for, if I know that it likely exists, and know the keywords to use to find it. But it’s much harder nowadays for sure.
I wonder how much influence google had in lowering the content quality over the years? After all, most SEO spam was a direct response to all the ludicrous requirements they've forced the whole web into, which eventually only SEO spam were willing to commit to.
I also wonder if google just stopped existing, would the web heal over time?
i have a radio that can "hear" down to -130dBm, i've proven this empirically. Cellular signals work at -12dB or more below the noise floor, wspr works even lower than that. Lightning is broadband noise, and yet i can still use digital stuff when there's lightning storms.
I don't buy the signal to noise argument. For example, whenever i get on youtube and get fed some content, i can immediately tell if it's had AI involved anywhere, and thumbs it down. I won't recommend it, i've called people out for linking such tripe to me (or others).
Hear me out - google got bad about 11 years ago when the dorking stopped being effective, right around the time of the spotlight search results and the sponsored junk taking the top results. Around this time, various agencies (news, etc) started gaming the SEO to respond to any remotely related search with whatever the news was currently. Google chose not to "fix" this, because we're not the customer. DDG was better for a few years for real results, too, but that has gone downhill as well.
The current zeitgeist uses stuff like tiktok and facebook for "web searches" - "food trucks near Austin, TX" or so. No one really uses web search like people on this site do, and google couldn't care less if we don't like the search results.
Is it actually more complex to provide good results, or is it just more profitable not to?
I have a hard time believing an organization like Google doesn't have the resources to provide a search engine that's just as usable as what they had 6 years ago (around the time I feel like the decay really set in). Seems a lot more likely that it's just more profitable to serve up garbage sponsored content.
Definitely more profitable not to. Especially as Google is an ad company, not a search company.
I’d rather see a world with numerous paid/subscription search engines, that are motivated to do nothing but return search results well. I expect you would see some of the SEO crap getting solved.
i cant remember where i read this, but something about how google ranks site that have google ads higher than sites that dont. makese sense, its evil, but makes sense, thats why we get all this scrapped spam. is there any more info on this?
this is like focusing on one single problem as being the cause of the decline of the United States. It's actually a lot of things combining and there's not going to be one fix
Intentionally ranking sites with Google ads higher would be a huge antitrust liability, so no way they're doing that.
On the other hand, they can achieve virtually the same outcome while keeping plausible deniability by just not doing anything that would downrank sites with ads (of which a significant chunk is likely to be Google's).
It doesn't need to be public to become an antitrust liability. Internal written material can still come up during discovery, potentially even in unrelated cases.
Therefore the safest option is to never openly discuss it or intentionally do it and instead use other means to achieve the objective (don't intentionally rank spam higher, just defund/cancel any projects that would make it rank lower).
> Internal written material can still come up during discovery, potentially even in unrelated cases.
Yup, and I think we've seen how careless and thoughtless Google is, as an organization, with internal comms in the Epic case. It would be shocking if it hadn't been discovered in that, or a prior, lawsuit.
The problem is that even if providers of the service are 100% trying to provide a great service, everyone on the web will always be min-maxing to appear on top.
To me it is only due to the ads, google and bing return nothing but ads on the first page. Plus for me to have the joy of seeing these ads, I need to got through a CAPTHA that I need to try multiple times.
I reckon these days search is pretty difficult and everyone knows how to game it. I recommend using a search engine that lets you effectively change which sites are shown. You can do this with Kagi, or with Google's Programmable Search Engines - I'm sure there are more too.
In particular I block Youtube, not because they aren't sometimes correct, but because I don't want videos polluting the regular results - it just takes too long to get info from videos.
An ability to upvote results for a given query seems tantalizing but I bet it would be gamed too. The DIY approach seems to be the only tractable one.
In my case I only only results from domains I believe are correct. The whitelist approach does have downsides. Usually I'll vet new potential domains through social means like Reddit and this site, rather than identifying them through the search results. I believe there's an inherent tradeoff between discoverability and the gameability of the results.
Though I do sympathize with folks who reminisce about 2008 Google Search results, there were probably orders of magnitude less content out there and a complete ignorance to how valuable your place is on your business and thus no SEO.
I also personally disagree that yt-dlp is the "correct" result for the average user when they search Youtube Download. I highly doubt the average user would know or care to use the command line. A website front end would be more actionable for them.
I'm a big fan of the non commercial site search engines because of the gaming aspect. If you're not generating revenue from the clicks the game mostly goes away.
I'm not saying people aren't entitled to make some money, but it clearly incentivizes user hostile behavior.
Maybe make it an option because legitimate sites like journalism also use this model.
It works not because they're somehow smarter or have more resources than Google at detecting spam/SEO, it's because unlike Google (and other ad-supported search engines), they make money from result quality and have an interest in blocking spam.
Google on the other hand makes money off ads (whether on the search results page itself or on the spam sites), so spam sites are at best considered neutral and at worst considered beneficial (since they can embed Google ads/analytics, and make the ads on the search results page look relatively good compared to the spam).
Black-hat SEO has been around since the early days of search engines and they managed to keep it at bay just fine. What changed isn't that there was some sudden breakthrough in malicious SEO, it's that it was more profitable to keep the spammers around than to fight them, and with the entire tech industry settling on advertising/"engagement" as its business model, the risk of competition was nil because competitors with the same business model would end up making the same decision.
The same reason is behind the neutering of advanced search features. These have nothing to do with the supposed war on spam/SEO, so why were they removed? Oh yeah because you'd spend less time on the search results page and are less likely to click on an ad/sponsored result, so it's against Google's interests and was removed too.
Google is a complex system so “want” can just include we are making money from the blog spam and while we don’t like it other things take priority over fighting it as effectively as we could.
It's never tinfoil-hat to assume that a corporation is, at very least, making sure not to fight too hard against any activity that brings it more revenue.
That's why taking the money out of the click is effective.
There can be other models for making money, but methods that really on casting a wide net and driving low quality traffic is the thing that shouldn't be indexed or at least labeled as such
But the author tried Kagi and the results don't appear to be noticeably different, filled with scammy adspam just like Google and Bing. Kagi's results seem to mostly aggregate existing search engines [1], so this isn't much of a surprise. Perhaps a subscription-based service that operates an index at Google's scale might help, but no such thing exists to my knowledge.
Right, but Kagi has built in tools to make it easy to fix that. Blocking those spammy sites from ever showing up again. Moving certain sites up the ranking, and so on. These features mean that over time my Kagi results have become nearly perfect for myself.
This is addressed in the article. As Hacker News readers and expert computer users, we have a bag of tricks that we can reach into in order to make our searches perform better. With a similar level of effort and an expert user's intuition you can get good results out of any search engine. Not so for the average user. In fact, again paraphrasing the article, Google's original claim to fame was that you didn't have to spend a lot of time doing exact keyword matching and fancy tricks in order to get good results.
> In particular I block Youtube, not because they aren't sometimes correct, but because I don't want videos polluting the regular results - it just takes too long to get info from videos.
Funnily enough, lately I've been prioritizing YT videos more when searching. So many sites now are just regurgitated SEO farms with minimal quality, and easy to see why: it's minimal effort to produce and cheap to host. But making a video takes time and effort, so has a much higher barrier to use as a click farm.
More than once when traditional search failed me, I went to YT and found some video from 2009 clearly and eloquently explaining what I'm looking for in detail, and without any distractions because the person authoring the video clearly didn't specialize in the media format or show interest in experimenting.
I've found it to also be a better source when looking a product to buy. Want to know which fan to get? Turns out there's a channel from a dedicated guy who keeps finding ways to test different fans and their utility and with multiple videos demonstrating his approach and findings. The mainstream channels aren't all that useful, but there's a ton of "old web" style videos (some even recent) passionately providing details for almost anything you'd think to search. And they're a gold mine.
Would a browser feature that skipped to the relevant parts of the video based on closed captioning and understanding search intent be useful? It seems like this would be a good way for Google to fight to stay relevant in UX vs having the chat bots just quickly spitting out a readable answer. Hunting through ad laden webpages is annoying. Seeking to the relevant section of the video is a solvable problem, especially for videos above some viewership threshold.
Google seems to be taking much more advantage of YouTube's transcription feature lately. The first addition was the (ok, gimmicky) animation on the Subscribe button when someone says the dreaded like. Hopefully a sign of things to come.
Overall AI summaries are very welcome for a certain subset of YouTube which is sadly dominated by sponsored, clickbait, and ad-driven content.
Didn’t Google try this already? It seems useful to me, at least. IMO the next frontier of search is not better hypertext, it’s podcasts, audio, and video.
> Seeking to the relevant section of the video is a solvable problem
...and it has already been solved, though partially: SponsorBlock allows people to add a "Highlight" section to a video, which denotes the part of the video which the user most likely wanted to see (sans the "what's up guys", "like and subscribe", etc.)
Of course, it's not perfect: it relies upon humans doing the work, though some may see that as a positive over something more computerized.
Do you have some tips for finding concise videos that answer the question you are asking? I am finding more and more obvious LLM bullshit in results, so I am willing to try some other tactics. But I am not ready to spend the minutes watching videos to see if it is actually relevant or a waste of time, always artificially long to increase ad revenue.
For me, it really depends on the type of video. For fixing cars, I'm usually looking for something specific enough that there isn't a lot of chaff. It was probably recorded and edited on a phone just to splice the clips together. Probably the default thumbnail that youtube extracted from the video.
For product videos, if Project Farm did it, look there first. Otherwise, I look for someone has a lot of videos for competing products with basically the same format, not over 10 minutes.
Tech videos are the hardest, I often still prefer text. Maybe look for links to the docs in the description? I still get duds though.
Wish I did, but here you're at the algorithm's mercy, unfortunately. One possibility is subbing/accruing watch time on channels that you find provide you the right value, so that the algorithm might recommend similar channels on other subject matters.
> But making a video takes time and effort, so has a much higher barrier to use as a click farm.
> The mainstream channels aren't all that useful, but there's a ton of "old web" style videos (some even recent) passionately providing details for almost anything you'd think to search. And they're a gold mine.
This won't be the case for long. YT is already starting to be polluted with spam and AI generated content, which will get more and more common. The same thing that happened to the web in text form, will happen to videos.
I think the only solutions are using allowlists for specific domains, and ironically enough more AI to filter specific results. Or just straight up LLMs instead of web search, assuming they're not trained on spam data themselves.
Yeah. I was recently looking for videos comparing two smartphones and among top ranked videos there were videos that just show the phones side by side and the video consists of showing specs side by side and videos that just have LLM-generated text, added to the video with TTS.
One critical difference is the date attached to youtube videos. It's easy to verify that a video was made before this tech was available, but you can't do that with websites, or search engine result pages.
It does limit utility for more modern needs, unfortunately.
That's curious, I generally hate video due to inability to glance over content, and the few attempts I made to actually find useful information I searched for resulted in... spammy extra low effort video content that did not answer my questions.
Depends on what you’re looking for. A blog post about how to play Search and Destroy by The Stooges is not as useful as a video of James Williamson himself showing you the riffs!
Well, I don't think I'd be able to learn much just from watching the concert: teaching is fundamentally different from doing.
So I think even that example does not universally hold. I'd still appreciate a write up with tips on what's important and if there are any transitions to focus on with only the bits on video where some of that is demonstrated.
Now, I can barely contort my fingers into one riff, so I lack the knowledge to understand what I am missing, but I'd still have a hard time learning that from video.
I’m not talking about concert footage, I’m talking about James breaking the song down and showing you the riffs at quarter speed.
Until a recent YouTube video I was playing the song incorrectly. It’s blazing fast and the mix is sort of insane so it’s very hard to hear exactly what is going on. And the tablature isn’t going to let you see how his body fits into the groove.
This is tacit knowledge we’re talking about, not book learning. Guitar instruction is always hands on.
Almost everything is hands-on (everything apart from things you really can't do hands-on, like exploring black holes): I don't remember seeing someone come out of reading a book on programming and being a master programmer.
But video is not hands-on any more so than text: if it was, live concerts and sports games and other performances would not be such a big deal. Sure, video is richer in some signals (audio/video), but poorer in others (introspection, pacing and focus...).
That does not mean I can't read to understand a new topic or to be prepared to look for subtleties in a hands-on performance.
If anything, to a great student, they should be complementary, but still, each student will have one or the other contribute more to their learning, and that depends both on the teacher, but also on the student.
I can’t wait until video transcripts get fed into LLMs just to eliminate the whole “This video is sponsored by something-completely-unrelated, more about them later. What’s up Youtube, remember to like, share, subscribe… 5 entire minutes pass on similar drivel… the actual thing you want, but stretched out to an agonizing length”
Usually people leave a "highlight" marker which tells you where you're supposed to jump to. Along with the regular "This video was brought to you by <insert>VPN".
> Though I do sympathize with folks who reminisce about 2008 Google Search results, there were probably orders of magnitude less content out there and a complete ignorance to how valuable your place is on your business and thus no SEO.
That was a decade after Google was created and people certainly understood SEO and Google was constantly updating its algorithm to punish people who were trying to game the algorithm.
The wikipedia page on "link farming" for example references it happening as early as 1999 and targeting SEO on inktomi:
I remember some internal presentations at Amazon around ~2004 about how boosting Google SEO on Amazon web pages increased traffic and revenue (and Amazon was honestly a bit behind-the-curve due to a kind of NIH syndrome).
I have a hard time believing it's so difficult for a search engine to distinguish between a credible, respected website that has been around a while with some generated garbage that exists to be a search result. We humans can tell them apart, so in principle, computers can too.
Yes, this should be table stakes for a classifier - a company with the resource of Google can definitely solve that problem if they weren't themselves in the business of spam (advertising) and benefited from spam sites (as they often include Google ads/analytics).
Always “table stakes”. Do you think in buzzwords also? I’ve always wondered this. Or do you think normal words and then translate it into this bandwagoning / membership proving garbage ?
We eventually have to ban accounts that post like this, so if you'd please stop doing that, we'd appreciate it. On HN the idea is: if you have a substantive point, make it thoughtfully; if not, please don't comment until you do.
Google's PSE is neat but there isn't a good way to manage switching between them. They could easily add a little dropdown to let you select which one to use as part of the public link UI they provide for each one individually. Giggle[1] gives me this ability and I run it locally (alongside Kagi) for more specific things to target domain lists I've been building over the years.
The issue with traditional search engines is that keyword-first algorithms are extremely gameable.
Try https://search.metaphor.systems - it's fully neural embeddings-based search. No keywords, only an embedding of what the actual content of a webpage is.
How is that different from keywords? Embeddings aren't magic, they're just page content. Content is trivial to game since it's controlled by the website owner.
edit: The results are also from my quick QA not that great. Searching for "what is the best mouse to buy" leads to links to buy random mice versus review summaries or online discussions on mice. One of the recommended queries of "Here is a great fun concert in San Francisco" leads to some really bizarre results in non-English languages that have nothing to do with either SF or concerts.
edit2: Also, Google has been using LLMs part of their search since at least 2018 so definitely not just keyword matching there.
Yup, definitely still gameable but if the model learns what high quality content is like and what high quality webpages there are (which it does), then the only way to game would be to be great :)
For your search - I would recommend turning autoprompt off and searching something like "Here is a great summary of the best computer mice to use:".
Our embeddings model is trained on how links are talked about on the Internet, if that helps with querying. So you have to query like how someone would refer to a link before sharing it
> Our embeddings model is trained on how links are talked about on the Internet, if that helps with querying. So you have to query like how someone would refer to a link before sharing it
So it's not high quality web pages but web pages that people talk about a lot which is expected since no one has an oracle that says what high quality is. The embeddings are merely a proxy and generalization for "how links are talked about on the Internet." That can be gamed at scale just like every other signal any popular search engine has been based off of.
The first result vtubego.com is a 144MB downloader app. The page contains "Pricing Plans
Lorem ipsum dolor sit amet, placerat verterem luptatum phaedrum vis, impetus mandamus id vix fabulas vim." above its 3 paid plans (there is no free plan).
I haven't installed the downloader app, so I'm not sure if it lets me download youtube videos for free.
The second result "ytder.com" is a redirect to "https://poperblocker.com/edge/" which seems to be a browser extension for Microsoft Edge that protects the user from the Holy See. I'm not using Edge and I'm trying to download a Youtube video.
The third result download-video.net says that it can download videos from a list of sites. Youtube is not in the list, but let's try anyway. If you put "https://www.youtube.com/watch?v=IkYVmtgxebU" into the text box and click "download" you get "500 SyntaxError: Unexpected token '<', ""
At this point I gave up, but please let me know if any of the results work.
> with Metaphor you'll get only Youtube downloaders
I clicked into the top 5 results, none of them were real youtube downloaders that worked, so I clicked the next 5 results, then I finally got one single (really slow) downloader that worked. 1 out of 10 top results
While I've made huge improvements to the algo recently, I do think Marginalia Search got a bit lucky with the sample queries, as it is still IMO far more hit and miss than many alternatives, but that also speaks for how hard evaluating search quality is.
Its efficacy is also strongly dependent on understanding that it's a keyword search engine with no semantic understanding.
I notice you completely avoid the question on how a single developer can do so well ;)
I do think that search has gotten much worse but my ability to know the magic words like “ublock origin” instead of “Adblock” and “yt-dlp” instead of “download YouTube” and phrase my search has gotten better.
We’ve all been doing prompt engineering against the Internet-wide LLM that is the spam houses.
> I notice you completely avoid the question on how a single developer can do so well ;)
As much as I enjoy the notion of somehow being a 10,000X developer, it's probably mostly that modern search is a filtering problem, and MS does filtering fairly well.
> [...] but that also speaks for how hard evaluating search quality is.
Would you be able to share some of your personal highlights regarding this?
I've partially kept up-to-date with the DIY, non-corporate search space (YaCY and friends). I'd love to understand a bit more behind the engineering decisions made when creating a search engine; it seems like a very hard problem to solve.
P.S. Marginalia is a very impressive piece of work, overall -- I've heard nothing but positive remarks from users on here. I've been meaning to try it for a while, but time constraints have... well, constrained, thus far.
I just tested Mariginalia and it was completely unable to lead me to a Wikipedia or imdb page when searching for "driver ryan gosling" and variations. It just listed lots of random articles.
That.. is kind of the point of this particular search engine.
> This is an independent DIY search engine that focuses on non-commercial content, and attempts to show you sites you perhaps weren't aware of in favor of the sort of sites you probably already knew existed.
Well that makes sense, but I wanted to push against the result that the OP seems to take away from their test, which was that Marginalia seems to work well for the common user.
There's also a known bug with Wikipedia in particular, I do index it but the results are never ranked particularly high. I haven't fixed it because I don't want Wikipedia to be the #1 result for every search. Feels like most people are aware of Wikipedia and don't need help finding it.
I often do a Google search, and then go directly to the Wikipedia result. My reasoning is that during the initial search, I don't know if there's a Wikipedia page about that topic, and I might need a fallback option.
Unless it's something related to medicine; then you have to explicitly add "wiki" to the query. Some public health thing to discourage hypochondriacs I guess, but it's very annoying.
I have a suggestion for the “About” section at the top of Marginalia’s landing page. I think it would read better like this:
> This is an independent DIY search engine that focuses on non-commercial content, and attempts to show you sites you perhaps weren't aware of [instead] of the sort of sites you probably already knew existed.
Showing one thing “in favor of” another seems contradictory in this case.
Just my feedback after trying to finally get to what it is exactly.
I tried to find marginalia on DDG, not on the first page. Google has it after some garbage.
If I go to marginalia.nu I get a SSL error. search.marginalia.nu works
If i search on marginalia for duckduckgo there first link is somewhat relevant but is about the app, all the other links are related to DDG but of curious relevance.
If I search for ublacklist mentioned above, I do not see anything directly relevant.
Hmm, what's your browser? I renewed the cert today... Only thing I can think of is that it might not like a wildcard cert for the bare marginalia.nu domain.
Hmm, can't reproduce it myself, but firefox has a nasty habit of quietly "repairing" these types of misconfigurations by redirecting from one subdomain to another. I've added marginalia.nu as a SAN, should hopefully work now.
It's tricky though. I think a lot of people think they want raw keyword search, but what they really want is a search experience that makes intuitive sense.
If you lean too much into embeddings and so on, it's easy to get errors that don't make sense to a human being. It's extremely frustrating when you experience "I typed X, why am I getting results about Y?!"
That said, I think there's a sweet spot with some magic, where it genuinely just makes search better. But it's like perfume, if it's immediately obvious that it's there, it's probably a fair bit too much.
The return of something like Yahoo Directory would be most welcome. There is great utility in having more than one approach into a data space. That we have been stuck with essentially one way in for over a decade means that there is a great deal out there which would be great to access but which has been rendered invisible.
I use serpapi for my hot RAG and the results are fine.
Brave search API is obscenely overpriced. I hope someone is working on Search because Google has become a singularly garbage company. Propping up DEI is sinful enough but just failing to compete is lame. /shrug
Do search engines censor political topics these days? If you search "truthsocial" on ddg, the truthsocial.com website is the first hit. But if you search "trump truthsocial", it doesn't give you trump's truthsocial page, and doesn't even give you truthsocial.com within the first few pages of search results.
Since ddg uses bing, does anyone know what is happening here at bing? It looks like google results are similar.
DuckDuckGo (and by extension perhaps Bing, assuming identical upstream results) has some terrible results when trying to filter by all kinds of domains.
There's a power tools review/news site that returns zero hits for the actual domain when searching its name (which is the same as its .com address). While for some domains even searching using the `site:` parameter will give far fewer results when paired with a query than just searching the domain name + query sans the TLD (the router firmware site openwrt.org is among such).
It's a mess and reporting it hasn't any difference ime in the past 3 years. So I'd be reluctant to say irrelevant results are due to censorship unless there was more evidence.
I have concluded that Google definitely censors search results relating to the Ukraine war, after vainly searching for articles about documented Ukrainian war crimes (reported in mainstream Western media like NYT/WaPo).
I'm not seeing this. I Googled "war crimes by ukrainian soldiers" and the top link was an Amnesty International Article, "Ukraine: Ukrainian fighting tactics endanger civilians".
You're right: I just checked and there are several hits for events that happened over a year ago that I couldn't find at all with Google back then. Shame on me for not checking before I posted. I have no idea what happened but apparently it's now fixed.
I noticed that the author uses ChatGPT3.5 rather than 4, which is a rather large difference. I don't have the knowledge to rerank all questions the author asked, but I will say that a test of ChatGPT 4 leads me directly to youtube-dl, which is better than every other search engine listed.
That was the first thing I checked reading the article. Although the argument would be 3.5 is free - any comparison of systems against ChatGPT that isn't using ChatGPT 4 can be dismissed almost out of hand; there is not much point talking about ChatGPT if it's not using ChatGPT 4 and making proper use of its capabilities.
That is not to say that there aren't valid criticisms of and shortcomings in ChatGPT 4 - just that it's not useful to say ChatGPT when it's referring to 3.5
Sure. Bear in mind I have custom instructions active - which, if you want to make full and proper use of ChatGPT, you should configure, along with customised GPTs - so I get lots of dot-point descriptions, because that's what I've asked for.
Also I would not normally write ChatGPT queries the same as I write them for search engines but for the sake of comparison, I'll use their queries verbatim except where my custom instructions affect the context too much.
I got - good results.
They got - "Very bad results (fails to return any kind of useful result)
ChatGPT: basically refuses to answer the question, although you can probably prompt engineer your way to an answer if you don't just naively ask the question you want answered".
Honestly, I have no idea if this is a good answer or not. But I don't use ChatGPT for answers that I don't have confidence that I can determine its veracity; if I needed to know this with certainty, I'd use ChatGPT as a jumping off point for my own research.
> Why do they keep making cpu transistors smaller?
> [Provide links to scientific sites that describe] why do they keep making cpu transistors smaller?
ChatGPT 3.5 was great, until 4 came out and now it is garbage in comparison.
But I suppose what I really want is for everyone who includes ChatGPT in comparisons to explicitly say which version they are using (and, if they are using 3.5 in their comparison I hope they at least try 4 first) and definitely not just say "ChatGPT" when they only mean 3.5. The difference really is that stark.
This is silly, most people aren't going to pay for ChatGPT, just like they won't pay for Google or DDG. So using 3.5 in this case is perfectly acceptable when we're talking about free software.
I also suspect as much, but obviously can't know for sure. IMHO it's intellectually lazy if not dishonest to benchmark against 3.5 and not make that fact clearly known upfront
A better benchmark would have had two entries for ChatGPT, showing both 3.5 and 4 results
I boggles the mind the extent to which people salivate over a system that cannot decide between a correct straight answer, something wrong but plausible, something wrong and impossible, or outright refusing to answer.
That's GPT 3.5. It sounds like you have a bit of an axe to grind with ChatGPT, but if you're going to do so, do try to grind it on the correct version.
The comment says it’s v4. Since there’s no information on the page either way (funny, considering the original complaint), I took them at their word. If you don’t believe them, that’s up to you.
For what it’s worth, I do have access to v4 and it did give me an answer right now. But since I also know even v4 can give you wildly different answers to the same question even if you ask them one right after another, that doesn’t prove it either way.
I actually find myself using bangs way more since I switched to Kagi from DDG. I think it's the AI bangs like !chat and !expert that got me in the habit of using bangs besides !g (which I never actually use anymore).
Pretty sure the reason Kagi is better isn't because they use multiple sources, it's just because they can use the presence of ads as a negative ranking signal, something that none of the major public search engines will ever do as it goes against their own business model.
I am not sure what the intention of this post is. In my handpicked results Kagi far outperforms Marginalia.
#1 "Gordon ramsey" (misspelled "Gordon Ramsay"). Marginalia shows "The Life I Imagine: are my cheeks red?". Kagi corrects to Gordon Ramsay and shows relevant results.
#2 "Ukraine war". Marginalia shows an article about the Russian Orthodox church and a Substack post about the war. Kagi shows Wikipedia, Al Jazeera, etc up-to-date summaries about the war.
#3 "Dildo". Top post on Marginalia is "Students for Concealed Carry Embraces UT Dildos | Students for Concealed Carry". Top posts on Kagi are Wikipedia (read) and Amazon (buy).
> How is Marginalia, a search engine built by a single person, so good?
I had a similar experience when testing Kagi after reading this. The top result for the “wider car tires” query on Kagi was a link to Physics StackExchange with some marginally informative answers [0], which would be easy to expand on in future searches. The second result was Reddit. Then a couple of incorrect/irrelevant pages but they don’t look like scams
Edit: I did just realize that I have StackExchange customized to be up-ranked. So that probably helps. But yeah, I guess this is why I usually get good results, which is something that generally still fails with Google for me.
I don't disagree with your assessment in full, but I don't exactly consider wikipedia and Amazon good results. Like they are big enough that if that's the result I want I can go to them directly. So like they aren't bad or wrong, but I can see the case for excluding them. Should something like Webster's dictionary be a top result?
I think for single word queries like that Wikipedia covers more ground than a dictionary. Personal preference, perhaps. If I need a definition I search for "define dildo" (Kagi shows Merriam-Webster, Oxford, etc dictionary entries).
Thanks! If you are that "single person" who built Marginalia... hope you are not taking my criticism personally. I am more annoyed by this blog post that uses a few handpicked queries to present generalized long winded conclusions that are completely disproven when using a different set of queries.
Yeah, its me, and to be fair I made a comment to a similar effect myself. Assessing search result quality is very hard, and this is definitely a pretty flattering selection of queries.
On the plus side - in addition to Marginalia's own success, you can take partial credit for how good Kagi search results are (IIRC Marginalia's index is one of the sources for Kagi search results). So... thank you for that!
It seems to me that the name "marginalia" is not just a random set of syllables. It sounds like it's doing what it says on the tin, which is gooder than not doing what it says on the tin. (distinct from whether what it says on the tin is something you want)
The appendix describing the individual search results is both entertaining and scary e.g.
"Two of the top three hits are how to install the extension and the rest of the top hits are how to remove this badware. Many of the removal links are themselves scams that install other badware."
I would love to see Perplexity.ai in the benchmark. It has completely replaced Google/DDG for information questions for me. I still use DDG when I want to do a navigational query (e.g. find the URL for a blog i partially recall the name).
While kagi was the product that most brought me joy in 2022, perplexity.ai has been the one for 2023, even though i only recently started using it. It's just been a joy to be able to iteratively discuss most of my searches.
I've been really enjoying Perplexity as well. It's a much better Internet/search focused experience than ChatGPT, Bing, or Bard. For anyone interested, until the new year (~20 more hours?) there's a code for 2mo free Pro: https://twitter.com/perplexity_ai/status/1738255102191022359 (more file uploads, choose your model including GPT4)
You can sync the settings and your personal blocklist to either Dropbox or Google Drive. It also has the ability to subscribe to blocklists. Mind, you need to manually turn on search engines and subscribe to lists. The uBlacklist subscriptions setting doesn't have any built-in feeds yet though. :(
uBlacklist is absolutely excellent: I've been using it for a few years now, with absolutely no problems.
Quick tip: turn on the 'Skip the "Block this site" dialog', and disable 'Hide the "Block this site" links' settings -- they make it much quicker to block spam websites (of which there are many on regular search engines).
Just today I was looking for an extension just to block Quora from search results. (Talk about a useless site that seems to uselessly outrank Wikipedia on google lately — what on earth is Google up to?) I’m thankful I saw your and your parent’s post.
When Quora was new I followed some topics, got to read interesting answers to interesting questions, but then some kind of enshittification happened. I've blocked it in Kagi now.
Yeah, for the the results of kagi are so much better than anything else, that it makes me wonder how objective can one be measuring search results.
I use google in a clients computer and it’s just horrible.
But it could also be a factor of the customizations I’ve made for my kagi. Ban quick a few paywalls sites, always put Wikipedia articles on top, prefer blogs than stackoverflow stuff…
I just tried it (free account), and it felt underwhelming, not many search results, or particularly interesting ones, for the image and video stuff I searched.
There was little to no spam, though, but not much to look at either. Maybe it might be useful when searching for stuff that usually has high amount of confusing spam, but otherwise not really useful for me...
The addon you linked (on the Firefox version) only requests permissions on google.* sites so I don't think it will work for DDG. Is there a separate extension, or am I misunderstanding something?
Blah blah blah. Could you lay this article out any worse? What are the queries you used to test? I want to try them too. Buried in here somewhere.
Using an adblocker is not expert anything.
That you've defined your own opinion for what some of the results should be blows the thing up.
Searching youtube downloader, many people would be fine with some of the ad covered but totally functional sites that pop up on Google. I use some of them every day for quick conversion tasks. I don't want any youtube-dl result. The average users don't either.
Download firefox? What's that? All the top links are fine? No one's looking at the 7th listing for a simple query to download a program.
Why do wider tires have better grip? .. what, sites like roadandtrack, prioritytire, reddit, some physics and stackexchange sites aren't good enough? they are.
The Vancouver snow report one also. Lots of major news sites. Some weathernetwork and almanacs. All totally acceptable results for a sort of variable question.
blah blah this is just a hate on for Google and a HN/nerd view of the world that the average user is nowhere near living in.
Completely agree. I personally thought searching "Vancouver snow report" to be extremely strange. Just search zip code or city name and weather. Two words. That's all you need to get results. What the hell is snow report? Do you even think you can trust weather reports 10+ days out?
For whatever it's worth, I think your comment would be a whole lot more convincing without its first and last lines, which had the effect of making you sound (at least to me) like you're shallowly dismissing the article.
Can someone tell me why Bing, and thus DDG, has switched to prioritizing local results? I'll search the most inane things, like lyrics to a song, and get results for local businesses containing maybe one word in common.
It's most frustrating with phone numbers. I picked up the habit of searching the random numbers that called me, to try and find out if they were possibly important. I used to get a bunch of spam sites that clearly existed to profit off me making those searches.
Both Google and DDG have removed those spam sites, even though they were useful at times. Google will tell me the number is in some random PDF that contains a few of the digits, then no other results. DDG will say the top result is my local police department, something that freaked me out the first few times.
Maybe it was an attempt to make better their results for local results?
When searching for results from my country in DDG (picking the country in the drop-down below the search box) still returned results from the USA or other countries. Even when searching stuff in the local language. Maybe they tried to fix that because it really sucked, so much I never used it again for searching into local websites.
This is the one area it still ignores my location. I live in a town named after a UK city, there's several bigger towns in the US with the same name. I just searched "McDonalds city name." I got results for the locations at least half the US away from me, as well as Uber Eats GB.
If you’re going to search for phone numbers you’ll want to ensure you enable verbatim searching under tools on Google, and put the number in quotes, perhaps in “xxx-xxx-xxxx” OR “(xxx) xxx-xxxx” forms. Many of the sites you mention are fake sites with fake contacts just for ad serving, and I’ve read in some few cases the scammers seeded the spoofed numbers they appear to call from on to the sites they control to see who googles their phone numbers.
Yeah, I've noticed this as well with DDG recently: even with the localised checkbox disabled it still prioritises them, which often is very frustrating as the results are then almost totally useless.
However, more generally, I've personally found that DDG (and maybe Bing's then?) localised results are just really bad, and have been for the multiple years I've been using DDG and it's had this feature: I'm in New Zealand, and enabling localised / region-based search still often provides results to pages with TLDs like "co.uk", ".ca" and ".pl" (these latter are really common for content-generated spam in my experience), which I just can't understand...
Unfortunately, I have found that Google's results are usually a lot better in terms of being "location-aware" than DDG, at least when that's what you want...
DDG is just repackaged Bing. Always has been. I remember looking into them when I was ready to job-hop many years ago, and they asked for dedication to their search engine as their foremost requirement for employment. It's the "drop-shipping" equivalent of search engines.
I really don't get that sentiment. Currently Kagi is just as dependent on Google as DuckDuckGo is on Bing. That might only be temporary of course and Kagi does seem to be working on a search engine of their own.
Rather than wanting Kagi to take the place of DuckDuckGo, it would would be better if Kagi could take users from Google, and then when ready, drop Google as a search provider.
I don't think they use Bing, but yes, Google, Marginalia, Yandex, Brave and others. I still fail to see how that's different to DuckDuckGo, who also run their own crawler. It's really weird that people are almost hating on DuckDuckGo for how they run their search engine, while applauding Kagi, for doing the same, but with a different business model.
You can buy access to the Google Search API, which is what I assume Kagi does. Building your product on being able to circumvent some Google restrictions seems like a bad business move, if you can buy the same service for a reasonable price.
DDG used to be the HN darling and you would get downvoted for saying anything negative or even insinuating that they are relying on Bing. Now the spot has been overtaken by Kagi but it looks like it suffers from the same problems. The counterargument that they have their own index as well is the same that was used for DDG, when the reality was that it was only used for widgets and other fluff. Let's see how it plays out for Kagi.
I’m confused, you are searching for, specifically, a local phone number and you are upset that the machine interprets that as you looking for a local result? That’s what most people expect from a local number search.
Perhaps the incorrect thing is not your internet search results, but actually your phone carrier for lying to you and telling you that a caller has a local number?
The number is local, and occasionally I've searched and found the number was a local clinic or business that had legitimate reason to call me but not leave a message. In those scenarios, close to all ten of the numbers are found on the page.
The top result being my local police department because it shares the same area code and has maybe one other number in common is clearly a bad result. It does this even if the phone carrier isn't lying to me and the caller does have a local number, like the increasingly common political spam calls.
Man, thank you for saying this. Stuffing results with geolocated local junk despite explicitly opting out by choosing “All regions” is so frustrating. This wasn’t happening a year or two ago. I submit negative feedback about it constantly, but I guess not enough people are doing that for anyone to notice or care.
I’ve also noticed a significant increase in attempts to stuff news into regular search results. I really do not appreciate being force-fed mental health poison. I don’t need it ever, but I especially don’t need it when I’m searching for some specific technical thing and then get emotionally sabotaged by some clickbait headline because … why? Some bullshit KPI? Why are tech companies so obsessed with pushing news into every orifice?
Nearly every local search is a leading indicator of buying intent and, therefore, is worth more money when served as a response instead of an authoritative response.
> Bing, and thus DDG, has switched to prioritizing local results
From what I can tell this is an issue with the Bing API that DDG uses that the DDG folks have been unable to resolve. I've tried many identical queries between DDG and Bing and while Bing does occasionally return incorrect local results, the completely irrelevant local results that appear on almost every DDG search do not seem to happen with Bing itself.
From what I understand, DDG is aware of the issue. I don't know why it isn't more of a priority.
Long time DDG user (>10 years) here, and it’s astounding to me that they haven’t prioritized making their own independent index to switch off Bing. I would have expected them to do it like 5 years ago, but there’s afaik no initiative to do so. It’s unfortunate and am now trying other engines like Brave search.
I also occasionally try Brave search when a DDG search fails. Sometimes Brave finds what I want, but I frequently get Captcha (and now proof of work) challenges that are quite annoying. I don't get this with any other search provider (though StartPage would frequently do this a while back). I hope this is just a phase, because I would likely use Brave Search more if not for this issue.
Thanks for confirming, we've just deployed some improvements to better handle VPN traffic and penalize legit users less. I hope this improves your QoL on Brave Search.
Have you tried perplexity.ai? It's like ChatGPT and Google had a baby. Looks very promising and I'm seeing a lot of tech leaders (example Toby of Shopify) moving to it.
No, FastGPT is GPT-2 based. I actually prefer FastGPT because its fast (duh!), and as it gives very concise answers and all the generated response carries footnotes with the link to the source.
I will admit that I can't read between lines here and just go ahead an ask: What does "bluesky thought leader" suppose to mean? (1) Any guesses who this may be? Why is he not quoted directly? (btw, the term is used 3 times, presumably to refer to the same person).
1: my reading is that this is a sarcastic denomination for someone that is supposed to be an innovation thought leader but actually is just defending the broken search landscape status quo.
Re: Kagi, I heard about it on HN, tried it for 100 searches, then subscribed. When I search for random JS and CSS things, MDN is the first result, and if it isn't, I can downrank whatever spammy site(s) are on top.
---
I wish I had a local LLM trained to detect clickbait and or low-effort content. I imagine searching YouTube and having all the clickbait collapsed together (just like Kagi condenses listicles), with the remainder being potentially high-quality content. Don't know how feasible this is right now.
Yeah. At first I primarily used Kagi to move away from Google as a company, hoping for results that were equally good. But Google search actually feels crappy now in comparison.
I keep Google Maps around for a similar reason; Apple Maps works well, but things like business hours are wrong often enough for me to double-check in Google Maps.
No for business info. Apple pulls directly from Yelp. issue is when you want more info or you want to get a closer look at the business and tap any of its images, it will take you straight to the app store to download Yelp.
492 comments
[ 2.9 ms ] story [ 311 ms ] threadAlso, the article tested Mwmbl as well, not mentioned in the title here.
Having said that, I’m usually still able to find what I’m looking for, if I know that it likely exists, and know the keywords to use to find it. But it’s much harder nowadays for sure.
I also wonder if google just stopped existing, would the web heal over time?
I don't buy the signal to noise argument. For example, whenever i get on youtube and get fed some content, i can immediately tell if it's had AI involved anywhere, and thumbs it down. I won't recommend it, i've called people out for linking such tripe to me (or others).
Hear me out - google got bad about 11 years ago when the dorking stopped being effective, right around the time of the spotlight search results and the sponsored junk taking the top results. Around this time, various agencies (news, etc) started gaming the SEO to respond to any remotely related search with whatever the news was currently. Google chose not to "fix" this, because we're not the customer. DDG was better for a few years for real results, too, but that has gone downhill as well.
The current zeitgeist uses stuff like tiktok and facebook for "web searches" - "food trucks near Austin, TX" or so. No one really uses web search like people on this site do, and google couldn't care less if we don't like the search results.
I have a hard time believing an organization like Google doesn't have the resources to provide a search engine that's just as usable as what they had 6 years ago (around the time I feel like the decay really set in). Seems a lot more likely that it's just more profitable to serve up garbage sponsored content.
I’d rather see a world with numerous paid/subscription search engines, that are motivated to do nothing but return search results well. I expect you would see some of the SEO crap getting solved.
On the other hand, they can achieve virtually the same outcome while keeping plausible deniability by just not doing anything that would downrank sites with ads (of which a significant chunk is likely to be Google's).
Spam sites often include ads.
Therefore the safest option is to never openly discuss it or intentionally do it and instead use other means to achieve the objective (don't intentionally rank spam higher, just defund/cancel any projects that would make it rank lower).
Yup, and I think we've seen how careless and thoughtless Google is, as an organization, with internal comms in the Epic case. It would be shocking if it hadn't been discovered in that, or a prior, lawsuit.
So it's inevitably going to become crap.
But all in all, a very good article
In particular I block Youtube, not because they aren't sometimes correct, but because I don't want videos polluting the regular results - it just takes too long to get info from videos.
An ability to upvote results for a given query seems tantalizing but I bet it would be gamed too. The DIY approach seems to be the only tractable one.
In my case I only only results from domains I believe are correct. The whitelist approach does have downsides. Usually I'll vet new potential domains through social means like Reddit and this site, rather than identifying them through the search results. I believe there's an inherent tradeoff between discoverability and the gameability of the results.
Though I do sympathize with folks who reminisce about 2008 Google Search results, there were probably orders of magnitude less content out there and a complete ignorance to how valuable your place is on your business and thus no SEO.
I also personally disagree that yt-dlp is the "correct" result for the average user when they search Youtube Download. I highly doubt the average user would know or care to use the command line. A website front end would be more actionable for them.
I'm not saying people aren't entitled to make some money, but it clearly incentivizes user hostile behavior.
Maybe make it an option because legitimate sites like journalism also use this model.
Their only remaining incentive is to be good enough that people keep paying for the service.
Google on the other hand makes money off ads (whether on the search results page itself or on the spam sites), so spam sites are at best considered neutral and at worst considered beneficial (since they can embed Google ads/analytics, and make the ads on the search results page look relatively good compared to the spam).
Black-hat SEO has been around since the early days of search engines and they managed to keep it at bay just fine. What changed isn't that there was some sudden breakthrough in malicious SEO, it's that it was more profitable to keep the spammers around than to fight them, and with the entire tech industry settling on advertising/"engagement" as its business model, the risk of competition was nil because competitors with the same business model would end up making the same decision.
The same reason is behind the neutering of advanced search features. These have nothing to do with the supposed war on spam/SEO, so why were they removed? Oh yeah because you'd spend less time on the search results page and are less likely to click on an ad/sponsored result, so it's against Google's interests and was removed too.
Super tinfoil hat to believe Google wants to send users to blog spam websites (e.g. beneficial to Google).
Anytime there is money to be made, there is an effectively infinite amount of people trying to game the system.
There can be other models for making money, but methods that really on casting a wide net and driving low quality traffic is the thing that shouldn't be indexed or at least labeled as such
[1] https://help.kagi.com/kagi/search-details/search-sources.htm...
Funnily enough, lately I've been prioritizing YT videos more when searching. So many sites now are just regurgitated SEO farms with minimal quality, and easy to see why: it's minimal effort to produce and cheap to host. But making a video takes time and effort, so has a much higher barrier to use as a click farm.
More than once when traditional search failed me, I went to YT and found some video from 2009 clearly and eloquently explaining what I'm looking for in detail, and without any distractions because the person authoring the video clearly didn't specialize in the media format or show interest in experimenting.
I've found it to also be a better source when looking a product to buy. Want to know which fan to get? Turns out there's a channel from a dedicated guy who keeps finding ways to test different fans and their utility and with multiple videos demonstrating his approach and findings. The mainstream channels aren't all that useful, but there's a ton of "old web" style videos (some even recent) passionately providing details for almost anything you'd think to search. And they're a gold mine.
Overall AI summaries are very welcome for a certain subset of YouTube which is sadly dominated by sponsored, clickbait, and ad-driven content.
...and it has already been solved, though partially: SponsorBlock allows people to add a "Highlight" section to a video, which denotes the part of the video which the user most likely wanted to see (sans the "what's up guys", "like and subscribe", etc.)
Of course, it's not perfect: it relies upon humans doing the work, though some may see that as a positive over something more computerized.
For product videos, if Project Farm did it, look there first. Otherwise, I look for someone has a lot of videos for competing products with basically the same format, not over 10 minutes.
Tech videos are the hardest, I often still prefer text. Maybe look for links to the docs in the description? I still get duds though.
> The mainstream channels aren't all that useful, but there's a ton of "old web" style videos (some even recent) passionately providing details for almost anything you'd think to search. And they're a gold mine.
This won't be the case for long. YT is already starting to be polluted with spam and AI generated content, which will get more and more common. The same thing that happened to the web in text form, will happen to videos.
I think the only solutions are using allowlists for specific domains, and ironically enough more AI to filter specific results. Or just straight up LLMs instead of web search, assuming they're not trained on spam data themselves.
It does limit utility for more modern needs, unfortunately.
So I think even that example does not universally hold. I'd still appreciate a write up with tips on what's important and if there are any transitions to focus on with only the bits on video where some of that is demonstrated.
Now, I can barely contort my fingers into one riff, so I lack the knowledge to understand what I am missing, but I'd still have a hard time learning that from video.
Until a recent YouTube video I was playing the song incorrectly. It’s blazing fast and the mix is sort of insane so it’s very hard to hear exactly what is going on. And the tablature isn’t going to let you see how his body fits into the groove.
This is tacit knowledge we’re talking about, not book learning. Guitar instruction is always hands on.
But video is not hands-on any more so than text: if it was, live concerts and sports games and other performances would not be such a big deal. Sure, video is richer in some signals (audio/video), but poorer in others (introspection, pacing and focus...).
That does not mean I can't read to understand a new topic or to be prepared to look for subtleties in a hands-on performance.
If anything, to a great student, they should be complementary, but still, each student will have one or the other contribute more to their learning, and that depends both on the teacher, but also on the student.
I can’t wait until video transcripts get fed into LLMs just to eliminate the whole “This video is sponsored by something-completely-unrelated, more about them later. What’s up Youtube, remember to like, share, subscribe… 5 entire minutes pass on similar drivel… the actual thing you want, but stretched out to an agonizing length”
Usually people leave a "highlight" marker which tells you where you're supposed to jump to. Along with the regular "This video was brought to you by <insert>VPN".
That was a decade after Google was created and people certainly understood SEO and Google was constantly updating its algorithm to punish people who were trying to game the algorithm.
The wikipedia page on "link farming" for example references it happening as early as 1999 and targeting SEO on inktomi:
https://en.wikipedia.org/wiki/Link_farm
I remember some internal presentations at Amazon around ~2004 about how boosting Google SEO on Amazon web pages increased traffic and revenue (and Amazon was honestly a bit behind-the-curve due to a kind of NIH syndrome).
Always “table stakes”. Do you think in buzzwords also? I’ve always wondered this. Or do you think normal words and then translate it into this bandwagoning / membership proving garbage ?
If you wouldn't mind reviewing https://news.ycombinator.com/newsguidelines.html and taking the intended spirit of the site more to heart, we'd be grateful.
Edit: unfortunately your account has been breaking the site guidelines in a lot of other places too—here are some recent examples:
https://news.ycombinator.com/item?id=38825624
https://news.ycombinator.com/item?id=38825543
https://news.ycombinator.com/item?id=38783196
We eventually have to ban accounts that post like this, so if you'd please stop doing that, we'd appreciate it. On HN the idea is: if you have a substantive point, make it thoughtfully; if not, please don't comment until you do.
The other question I have is how long do these garbage results stay up for a particular query on average?
1. https://github.com/dan-lovelace/giggle
And changing the query to "ad blocker" like Google suggested raised ublock origin way up in the results
Try https://search.metaphor.systems - it's fully neural embeddings-based search. No keywords, only an embedding of what the actual content of a webpage is.
So in the mentioned example of searching for Youtube downloaders, with Metaphor you'll get only Youtube downloaders (https://search.metaphor.systems/search?q=This%20is%20the%20b...)
Full disclosure - I work there :p
What prevents websites from gaming their embedding? Switching to a similarity search doesn't prevent the results from being gamed.
edit: The results are also from my quick QA not that great. Searching for "what is the best mouse to buy" leads to links to buy random mice versus review summaries or online discussions on mice. One of the recommended queries of "Here is a great fun concert in San Francisco" leads to some really bizarre results in non-English languages that have nothing to do with either SF or concerts.
edit2: Also, Google has been using LLMs part of their search since at least 2018 so definitely not just keyword matching there.
For your search - I would recommend turning autoprompt off and searching something like "Here is a great summary of the best computer mice to use:".
Our embeddings model is trained on how links are talked about on the Internet, if that helps with querying. So you have to query like how someone would refer to a link before sharing it
So it's not high quality web pages but web pages that people talk about a lot which is expected since no one has an oracle that says what high quality is. The embeddings are merely a proxy and generalization for "how links are talked about on the Internet." That can be gamed at scale just like every other signal any popular search engine has been based off of.
Definitely excited to see how it holds up to daily use.
So far it gave me exactly what I wanted at the top for all of my test queries that were well formed.
As for asking “ignorant” questions both your service and the goog failed where phind gave me an actionable starting point (after a prodding follow up question: https://www.phind.com/search?cache=hmul4znpn7y4ei6qa64fosmc )
“max-height like css property for top and left”
Unsure if this sort of thing is even a goal of your project, but you won over a new user.
Wish you and your team all the best.
For paywalls/login - we play pretty straight, always obey robots.txt, etc.
Auto-prompted to: "Here's a helpful website for downloading YouTube videos:"
Also, this result is horrible:
“What does it mean if someone is not covered in nfl football?”
I haven't installed the downloader app, so I'm not sure if it lets me download youtube videos for free.
The second result "ytder.com" is a redirect to "https://poperblocker.com/edge/" which seems to be a browser extension for Microsoft Edge that protects the user from the Holy See. I'm not using Edge and I'm trying to download a Youtube video.
The third result download-video.net says that it can download videos from a list of sites. Youtube is not in the list, but let's try anyway. If you put "https://www.youtube.com/watch?v=IkYVmtgxebU" into the text box and click "download" you get "500 SyntaxError: Unexpected token '<', ""
At this point I gave up, but please let me know if any of the results work.
I clicked into the top 5 results, none of them were real youtube downloaders that worked, so I clicked the next 5 results, then I finally got one single (really slow) downloader that worked. 1 out of 10 top results
Its efficacy is also strongly dependent on understanding that it's a keyword search engine with no semantic understanding.
I do think that search has gotten much worse but my ability to know the magic words like “ublock origin” instead of “Adblock” and “yt-dlp” instead of “download YouTube” and phrase my search has gotten better.
We’ve all been doing prompt engineering against the Internet-wide LLM that is the spam houses.
As much as I enjoy the notion of somehow being a 10,000X developer, it's probably mostly that modern search is a filtering problem, and MS does filtering fairly well.
Would you be able to share some of your personal highlights regarding this?
I've partially kept up-to-date with the DIY, non-corporate search space (YaCY and friends). I'd love to understand a bit more behind the engineering decisions made when creating a search engine; it seems like a very hard problem to solve.
P.S. Marginalia is a very impressive piece of work, overall -- I've heard nothing but positive remarks from users on here. I've been meaning to try it for a while, but time constraints have... well, constrained, thus far.
> This is an independent DIY search engine that focuses on non-commercial content, and attempts to show you sites you perhaps weren't aware of in favor of the sort of sites you probably already knew existed.
I have a suggestion for the “About” section at the top of Marginalia’s landing page. I think it would read better like this:
> This is an independent DIY search engine that focuses on non-commercial content, and attempts to show you sites you perhaps weren't aware of [instead] of the sort of sites you probably already knew existed.
Showing one thing “in favor of” another seems contradictory in this case.
I tried to find marginalia on DDG, not on the first page. Google has it after some garbage. If I go to marginalia.nu I get a SSL error. search.marginalia.nu works
If i search on marginalia for duckduckgo there first link is somewhat relevant but is about the app, all the other links are related to DDG but of curious relevance.
If I search for ublacklist mentioned above, I do not see anything directly relevant.
Good. I love keyword search.
"Semantic understanding" can be so biased and ... just shady sometimes.
If you lean too much into embeddings and so on, it's easy to get errors that don't make sense to a human being. It's extremely frustrating when you experience "I typed X, why am I getting results about Y?!"
That said, I think there's a sweet spot with some magic, where it genuinely just makes search better. But it's like perfume, if it's immediately obvious that it's there, it's probably a fair bit too much.
Anyway, here’s Kagi’s bangs:
https://help.kagi.com/kagi/features/bangs.html
You can also make your own bangs.
That said, my point was that the bang directory has a bunch of the most useful sites in each category.
Brave search API is obscenely overpriced. I hope someone is working on Search because Google has become a singularly garbage company. Propping up DEI is sinful enough but just failing to compete is lame. /shrug
Since ddg uses bing, does anyone know what is happening here at bing? It looks like google results are similar.
- "truthsocial trump" works
- "trump truthsocial" doesn't work
There's a power tools review/news site that returns zero hits for the actual domain when searching its name (which is the same as its .com address). While for some domains even searching using the `site:` parameter will give far fewer results when paired with a query than just searching the domain name + query sans the TLD (the router firmware site openwrt.org is among such).
It's a mess and reporting it hasn't any difference ime in the past 3 years. So I'd be reluctant to say irrelevant results are due to censorship unless there was more evidence.
https://www.amnesty.org/en/latest/news/2022/08/ukraine-ukrai...
I use Google as little as possible because I don't like surveillance advertising but fair is fair.
That is not to say that there aren't valid criticisms of and shortcomings in ChatGPT 4 - just that it's not useful to say ChatGPT when it's referring to 3.5
Also I would not normally write ChatGPT queries the same as I write them for search engines but for the sake of comparison, I'll use their queries verbatim except where my custom instructions affect the context too much.
> download youtube videos
https://chat.openai.com/share/3e18e4f0-5527-4479-8a2f-ef17bd...
I got - good results. They got - "Very bad results (fails to return any kind of useful result) ChatGPT: basically refuses to answer the question, although you can probably prompt engineer your way to an answer if you don't just naively ask the question you want answered".
> [What] ad blocker [can I use?]
https://chat.openai.com/share/e1985d7a-c89f-4b5e-bb59-70bd11...
Looks good to me
> download firefox
https://chat.openai.com/share/3a62e5ae-8dbd-4179-8eb0-cc38ee...
Also good
> Why do wider tires have better grip?
> [Provide links to scientific sites that describe] why wider tires have better grip?
https://chat.openai.com/share/8cbcd1dc-b23f-41f3-83ad-f43f3d...
Honestly, I have no idea if this is a good answer or not. But I don't use ChatGPT for answers that I don't have confidence that I can determine its veracity; if I needed to know this with certainty, I'd use ChatGPT as a jumping off point for my own research.
> Why do they keep making cpu transistors smaller?
> [Provide links to scientific sites that describe] why do they keep making cpu transistors smaller?
https://chat.openai.com/share/dbb97ac0-840c-402c-a917-657af6...
> vancouver snow forecast winter 2023
> Environment Canada winter 2023
https://chat.openai.com/share/aab017d7-f86b-49c9-b5c0-86a0b1...
I don't know if almanac.com is any good but giving it the specific "Environment Canada winter 2023" query gave the expected very good result.
I think ChatGPT 4 generally provided very good results for the test queries, if you tailor the queries just slightly for the format
Does everyone or even most use ChatGPT 4? The most used version is -of course- by far the most relevant.
But I suppose what I really want is for everyone who includes ChatGPT in comparisons to explicitly say which version they are using (and, if they are using 3.5 in their comparison I hope they at least try 4 first) and definitely not just say "ChatGPT" when they only mean 3.5. The difference really is that stark.
A better benchmark would have had two entries for ChatGPT, showing both 3.5 and 4 results
And yet to other people it starts rambling about how that’s wrong and you shouldn’t do it and doesn’t give a usable answer.
https://news.ycombinator.com/item?id=38822040
I boggles the mind the extent to which people salivate over a system that cannot decide between a correct straight answer, something wrong but plausible, something wrong and impossible, or outright refusing to answer.
For what it’s worth, I do have access to v4 and it did give me an answer right now. But since I also know even v4 can give you wildly different answers to the same question even if you ask them one right after another, that doesn’t prove it either way.
From what I understand, it aggregates results from multiple sources rather than having their own indexer.
The results aren’t really any better, but the lack of ads and videos in the results makes for a cleaner experience.
I also haven’t yet taken advantage of the extra features to block certain websites from results.
Personally, I pay the $5 mostly in an attempt to support another competitor in the space.
Start using bangs, lenses and customized results ASAP, that makes a big difference.
For both the tire question and with respect to a youtube dowloader, the first results were on the nose with the addition of site:edu on Google.
Why this is needed and whether a noncommercial, information rich web portal should exist are questions for another thread.
#1 "Gordon ramsey" (misspelled "Gordon Ramsay"). Marginalia shows "The Life I Imagine: are my cheeks red?". Kagi corrects to Gordon Ramsay and shows relevant results.
#2 "Ukraine war". Marginalia shows an article about the Russian Orthodox church and a Substack post about the war. Kagi shows Wikipedia, Al Jazeera, etc up-to-date summaries about the war.
#3 "Dildo". Top post on Marginalia is "Students for Concealed Carry Embraces UT Dildos | Students for Concealed Carry". Top posts on Kagi are Wikipedia (read) and Amazon (buy).
> How is Marginalia, a search engine built by a single person, so good?
Because it's not good?
[0]: https://physics.stackexchange.com/questions/29903/why-do-peo...
Edit: I did just realize that I have StackExchange customized to be up-ranked. So that probably helps. But yeah, I guess this is why I usually get good results, which is something that generally still fails with Google for me.
That's bad if you're looking for a simple answer or basic fact, and good if you're looking for a few hours of reading.
"Two of the top three hits are how to install the extension and the rest of the top hits are how to remove this badware. Many of the removal links are themselves scams that install other badware."
EDIT: here's a search for tire (I don't know anything about tire, so maybe there's much better links out there, but this is pretty much what I was expecting. Not an ad or SEO in sight.) https://www.perplexity.ai/search/tire-3iuI9T6BQUSvu2tAhgsRmA...
https://addons.mozilla.org/en-US/firefox/addon/ublacklist/
https://chromewebstore.google.com/detail/ublacklist/pncfbmia...
You can sync the settings and your personal blocklist to either Dropbox or Google Drive. It also has the ability to subscribe to blocklists. Mind, you need to manually turn on search engines and subscribe to lists. The uBlacklist subscriptions setting doesn't have any built-in feeds yet though. :(
edit: THere are some feeds on the uBlacklist site though. https://iorate.github.io/ublacklist/subscriptions
edit edit: Found an even better list of feeds. https://github.com/quenhus/uBlock-Origin-dev-filter#other-fi...
Quick tip: turn on the 'Skip the "Block this site" dialog', and disable 'Hide the "Block this site" links' settings -- they make it much quicker to block spam websites (of which there are many on regular search engines).
I use google in a clients computer and it’s just horrible.
But it could also be a factor of the customizations I’ve made for my kagi. Ban quick a few paywalls sites, always put Wikipedia articles on top, prefer blogs than stackoverflow stuff…
There was little to no spam, though, but not much to look at either. Maybe it might be useful when searching for stuff that usually has high amount of confusing spam, but otherwise not really useful for me...
Kagi really shines when you are doing a standard search, though, which is what most people do most of the time.
And it seems like they also have a 1000 domain limit?
If the topic has ever come up the discussion and links are likely to be more relevant and better than your avg. wiki article
Using an adblocker is not expert anything.
That you've defined your own opinion for what some of the results should be blows the thing up.
Searching youtube downloader, many people would be fine with some of the ad covered but totally functional sites that pop up on Google. I use some of them every day for quick conversion tasks. I don't want any youtube-dl result. The average users don't either.
Download firefox? What's that? All the top links are fine? No one's looking at the 7th listing for a simple query to download a program.
Why do wider tires have better grip? .. what, sites like roadandtrack, prioritytire, reddit, some physics and stackexchange sites aren't good enough? they are.
The Vancouver snow report one also. Lots of major news sites. Some weathernetwork and almanacs. All totally acceptable results for a sort of variable question.
blah blah this is just a hate on for Google and a HN/nerd view of the world that the average user is nowhere near living in.
They are if the first six results are SEO bullshit. Which is the de-facto state of affairs for Google today: advertising traipsing around as search.
Whole article is rambling and silly and assuming.
It's most frustrating with phone numbers. I picked up the habit of searching the random numbers that called me, to try and find out if they were possibly important. I used to get a bunch of spam sites that clearly existed to profit off me making those searches.
Both Google and DDG have removed those spam sites, even though they were useful at times. Google will tell me the number is in some random PDF that contains a few of the digits, then no other results. DDG will say the top result is my local police department, something that freaked me out the first few times.
Query: “I’m coming out of my cage…”
Result (Ad): “You’ll be doing just fine with these amazing year-end closeout prices at Al’s Discount Car Barn. Gotta come down—you’ll want it all!”
When searching for results from my country in DDG (picking the country in the drop-down below the search box) still returned results from the USA or other countries. Even when searching stuff in the local language. Maybe they tried to fix that because it really sucked, so much I never used it again for searching into local websites.
However, more generally, I've personally found that DDG (and maybe Bing's then?) localised results are just really bad, and have been for the multiple years I've been using DDG and it's had this feature: I'm in New Zealand, and enabling localised / region-based search still often provides results to pages with TLDs like "co.uk", ".ca" and ".pl" (these latter are really common for content-generated spam in my experience), which I just can't understand...
Unfortunately, I have found that Google's results are usually a lot better in terms of being "location-aware" than DDG, at least when that's what you want...
You can report them: https://ised-isde.canada.ca/site/canada-anti-spam-legislatio...
Rather than wanting Kagi to take the place of DuckDuckGo, it would would be better if Kagi could take users from Google, and then when ready, drop Google as a search provider.
It's been available for ages. We used it to power the company internal search for a large enterprise I worked at 17 or 18 years ago.
https://help.kagi.com/kagi/search-details/search-sources.htm...
Perhaps the incorrect thing is not your internet search results, but actually your phone carrier for lying to you and telling you that a caller has a local number?
The top result being my local police department because it shares the same area code and has maybe one other number in common is clearly a bad result. It does this even if the phone carrier isn't lying to me and the caller does have a local number, like the increasingly common political spam calls.
I’ve also noticed a significant increase in attempts to stuff news into regular search results. I really do not appreciate being force-fed mental health poison. I don’t need it ever, but I especially don’t need it when I’m searching for some specific technical thing and then get emotionally sabotaged by some clickbait headline because … why? Some bullshit KPI? Why are tech companies so obsessed with pushing news into every orifice?
From what I can tell this is an issue with the Bing API that DDG uses that the DDG folks have been unable to resolve. I've tried many identical queries between DDG and Bing and while Bing does occasionally return incorrect local results, the completely irrelevant local results that appear on almost every DDG search do not seem to happen with Bing itself.
From what I understand, DDG is aware of the issue. I don't know why it isn't more of a priority.
Are you by any chance using a VPN while using Brave Search? (ProtonVPN?)
Thanks for your help, we're working on ways to reduce the number of captchas shown to VPN users and your feedback is very useful.
1: my reading is that this is a sarcastic denomination for someone that is supposed to be an innovation thought leader but actually is just defending the broken search landscape status quo.
---
I wish I had a local LLM trained to detect clickbait and or low-effort content. I imagine searching YouTube and having all the clickbait collapsed together (just like Kagi condenses listicles), with the remainder being potentially high-quality content. Don't know how feasible this is right now.
I do have to dump into google for local searches every once in awhile, but otherwise happy with it.