I just want an effective boolean search. I don't want it automatically fixing type-os or including synonyms in results. If I remember a specific phrase, or misspelling from some random blog, I want to type that in and find it. Google search has gotten significantly worse since they started trying to be a generalized question/answer tool. Of course it doesn't help that most of the internet is walled off now. I suspect that AI search will be very useful for looking up information, but awful for searching webpages.
Fixing typos in "search" illustrates the difference between "search" as a means to find out whether something exists in a database versus a means for an advertising company to direct the searcher to the most optimum page for supporting advertising services.1
What is the difference between showing someone the winning advertisement versus showing them the winning "search result".1 The advertising company makes a decision according to secret criteria which ad/search result to show. A long, successful tradition of using alphabetical or chronological sorting for information, objective criteria, is jettisoned in favour of manipulation for commercial benefit of a middleman. The advertising company gets to decide what comes first and last, according to subjective criteria. Then make it infeasible to see what is last.
Google has admitted that in the majority of cases, an appropriate ad does not exist to pair with a given search query. That is what I call the noncommercial web. It has little direct value to "tech" companies and advertisers. Yet "tech" company intermediaries want to surveil and control our access to it nonetheless. As the parent states, "most of the internet is walled off."
In sum, the advertising company, and "AI", allow people to search for web content that exists, cf. allowing people to search for whether certain web content exists. The advertising company will make "suggestions". However not all content that exists is made searchable by the advertising company. If the content has no commercial value, e.g., it is "unpopular", it is considered garbage.3
1. When we search Wikipedia, typos do not result in "guessing" by an advertising company. If something does not exist, Wikipedia does not try to guess "what we want".
2. The term "winning" applies because the advertising company makes the process an actual auction or an effective contest.
3. If the web is mostly garbage, then its users should be entitled to know that through discovery. It seems as if the discoverable web, the one optimised for the advertising company, is mostly commercial garbage anyway. It is difficult to be serious about "AI" when it is trained on mostly garbage.
Fixing typos and ordering results by relevance instead of recency enhance user experience. They also help serve better ads, but do you really want the top result to be the page that was first indexed 3 minutes ago?
Perhaps this illustrates the importance of defaults. Consider which defaults favour advertising.
Search Wikipedia for "exmaple.com", a common error when typing the string "example.com". By default, the results are the pages that contain this exact string.
Now try the same query with Google. To get pages with this exact string, one needs to enclose the string in quotes. By default the results will not be pages that contain the string.
"Tech" companies want people to believe "search" and "suggestions" (non-objective ranking, insertion of "popular" items, insertion of the search provider's own web properties, etc.) are the same thing. I have no issue with suggestions. Sometimes, but not all times, suggestions are helpful. However I do have an issue with people believing search and suggestions are the same thing.
IIRC, there was a point in the past where Google was ignoring quotes and there were some HN threads about the subject.
In truth the Wikipedia search results do not contain the exact string "exmaple.com". It appears Wikipedia drops the .com and queries for the string "exmaple", in both singular and plural forms.
It's hard to articulate, but especially in the past year I've noticed that Google (the search engine) has become a place that serves the advertisers more than the users. Adding more and more ads to the top of the feed, doing effectively nothing to combat SEO spam (that generally has AdSense on the site), etc.
Kagi comparatively feels like a search engine that serves me as the user. It's a hard difference to spot, but the results _feel_ higher quality because I've curated them to suit what I actually want to see when I search the web.
If there's a site that's unreadable due to adblock-block nonsense or is generally very low quality, I can block it and move on. If there's something that generally has good content, I can raise it to prioritize it in results, or if it's something that I know I'll want to see a lot of (like AWS documentation when relevant), I can pin it to the top of the feed.
> It's hard to articulate, but especially in the past year I've noticed that Google (the search engine) has become a place that serves the advertisers more than the users.
Not hard at all to articulate... Google is not a search engine, but an ad platform.
I'm in the opposite camp. One thing Kagi did for me was to remove that mental barrier you're talking about — I'm now okay with paying for a search engine if it keeps me from being the product. My problem is that lately Kagi has felt a bit like Google in that no matter what I'm searching for, retailers and low-quality blogspam seems to be at the top of the results. At this point, if there was another paid search engine, I'd probably jump over to that one.
Fortunately, at least some of those cases are solvable on Kagi using the domain-level ranking overrides - so if the blogspam is coming from the same set of domains, just set their ranking override to "Lower" or "Block", and/or set the legitimate results (like Wikipedia or StackOverflow) to "Raise" or "Pin".
The thing I don't understand is... how come this is even a selling point? This is an obvious feature that was loudly demanded by people since time immemorial. It's something you realize you want five minutes into using your first search engine. Yet for some reason, neither Google nor Bing nor DDG nor literally anyone else I know of has ever implemented it. It took until ~2020 for Kagi to do it, and for me it's ~40% of reason I'm a paying user (the other ~40% is because they implemented a feature I proposed on their feedback board: the ability to rewrite result URLs via regular expressions - something I wanted to avoid having to use an extension to turn "reddit.com" results into "old.reddit.com" automatically).
Since before Kagi existed I use a localhost-bound forward proxy to turn reddit.com into old.reddit.com and a variety of other tasks. I do not use popular browsers much for recreational browsing which is probably why this works so well for me.
Right, there was always some way to do it. I wanted to have this in the search engine, and I'm happy Kagi implemented it, because 1) I feel this is where this functionality logically belong, and 2) it's automatically available at every machine I use, including mobile devices, without requiring any extra work.
2) is what makes browser extension route a no-go to me: too much hassle in getting an "unlocked" Firefox on Android.
There is an extension I used to use which did it for Google, though it's been long enough that I don't remember what it's called. I think that it is an okay workaround, but if I need to trim out this kind of stuff, I consider it a problem with the underlying system. It would be nice to be able to import or subscribe to someone else's list of domain-level overrides. But at a certain level, if you're subscribing to Kagi, you're probably looking to avoid that stuff by default, and you should just not have to deal with it in the first place.
For most techies in the west at least the price of Kagi should be next to nothing.
I personally will happily pay $10 a month for a search engine that works and takes search quality seriously and has its incentives aligned with my best interest.
And no, Google, Bing and DDG are utterly broken for many of the searches I do every week and I have written about it a number of times here and elsewhere since 10 years ago.
I like paying because I know my interests as the user are Kagi’s priority. With these “free” solutions of course the priority is the advertisers which is often at conflict with users.
I'm a happy paying user... I almost never have to !g. I tried DDG before kagi, and it just seemed like a worse version of google to me. I definitely prefer the kagi results.
"THE COLLECTIVE LIABILITY OF Kagi AND THE INDEMNIFIED PARTIES UNDER THIS AGREEMENT WILL NOT EXCEED $500 (FIVE HUNDRED DOLLARS)."
Why not limit liability to amounts received under the agreement. If the customer spends US$1000 on searches, then Kagi breaches the agreement, the customer can only recover "$500". WTF. Why is this so one-sided. That's left as a question for the reader.
There have been attempts in the past to insulate web searchers from Google's data collection but I do not recall anyone charging fees, openly engaging in data collection themsleves and asking paying customers to agree to a ridiclously low limitation of liability.
In these cases, one was trusting a potentially random person running a proxy, for lack of a better term. However no account, no credit card or other payment was required.
But that is not what Kagi offers. It requires disclosing one's identity via payment and account creation. It requires assenting to data collection. No way to opt-out. And to top it off, it requires agreeing to an pernicious limitation of liability.
In theory, having a paid agreement with a search engine website operator seems like it could be useful. The customer could, e.g., have an enforceable agreement with the website operator that prohibits the operator from certain activities, e.g., collecting data about the customer's use of the website and using it for financial gain or in a manner detrimental to the customer. If the operator breaches the agreement, then she can sue the operator.
What does Kagi actually promise to do or not do under this agreement. It promises:
"We will be good stewards of any personal information you share with us and we promise not to share your data with anyone else in any way, shape or form. Kagi's entire business is funded by its users and Kagi has no intention or interest in manipulating or monetizing user information in any way."
Why is "your data" not defined. For example, data that Kagi and its partners collect from your use of Kagi is arguably not "your data", it's Kagi's data. If it's Kagi's data then it can be shared with anyone "in any way, shape of form." Further, why is "user information" not defined. Is data collected about how people use Kagi considered "user information". If not, then it can be manipulated and monetised in every way.
Beyond the two statement about being good stewards, this entire "policy", like most "tech" company privacy policies, only states what the operator already does or does not do. At the same time it makes no warranties, so any statements about what it does or does not do could be grossly misleading or even false. There could be, and almost certainly is, other relevant information about what it does with the data it collects that is not disclosed.
A common tactic is to claim data is only used to improve the "product" or "service". Yet it is never disclosed exactly how this data is used to do that. What if the customer disagrees that such usage is actually improving the product/service. It leaves enormous discretion to the operator to use collected data however it sees fit. The customer cannot grant or deny consent for using data in some specific instance. She is not even informed about the instance, e.g., "We used would like to use your data to do X."
Other than the statement above about stewardship and keeping data for itself, does the operator does make any promises that it will do or refrain from doing anything in the future.
"We do X" is not a promise. "We will do X" is a promise.
How does the operator satisfy its performance under this agreement. For example, what if the website stops working. The only thing it has agreed to do is "be good stewards" of data it collects, whatever that means. Performance is satisfied by being god stewards. Nevermind actually providing a product or service. It appears the goal of the operator is to collect data. The "product" is only a means by which to do so. This collection will be financed by user fees.
As the GoogleSharing and Scroogle examples demonstrate, it is possible to provide web search without collecting data.
Google search can’t do what you want because there’s just too much noise in the “crawl everything” internet and the noise grows exponentially faster than the techniques that can automatically organize and rank it. There was a period of time where the ratio between indexing techniques and content worked, but we’re leaving that era now. All those things you hate about modern Google are necessary survival strategies for this model.
AI search can’t solve that problem. What it can do is give up on the idea of surfacing specific sources in favor of synthesizing answers from slurry of all the content, finding (or hallucinating) signal in the noise.
If you want strict keywords and boolean operators, you can’t expect to find that from the universal search engines that you’re used to. You want to look for and support curated engines that aggressively restrict what they index so that it contains high signal and low noise. Internal walled garden searches do this. Marginalia does this. Others probably do too. That’s what you’re looking for.
The trouble is there's no way for me to know how true this noise theory is. I'm given no option by these search engines to "just show me all the crap." What search engines do today is so filtered and heuristical that there are apparently phrases and ideas that have never been expressed on the internet ever, or at least that's the impression I'd get if I was more naive.
Besides, it's not like there isn't noise anyway. Ever click on the top results only to find that nothing matches your query and probably never did?
Google is not a search engine. It is a curation service. It only searches this collection of data that it decided was worthy of curation.
Google tells us this is for the best. We are forced to take their word for it since their algorithms are not public. Google tells us this is also for the best.
There is a way, just start spending millions/billions of dollars to index the internet...
Just figure out how not to index the endless content generation scripts, and places like Reddit that are 100s of millions of pages themselves. Or the content farms that are just copies of other sites.
> Google search can’t do what you want because there’s just too much noise in the “crawl everything” internet and the noise grows exponentially faster than the techniques that can automatically organize and rank it.
Go check search.marginalia.nu. Do some sample searches about history or linux or git or something and see how it often outclasses Google.
Then ponder the fact that it is run by one Swede and it runs on one (admittedly beefy but nothing unusual) workstation in his living room.
Personally at least I have seen enough to have made up my mind that it isn't impossible to do better than Google, they just don't want to.
I clearly didn't read your whole post. Have my upvote.
That said I think my point still stands: If 1 Swede and 1 server can do a much better job than Google in significant niches such as history, Linux and Git, then maybe a hundred Swedes and 10 000 servers can outcompete Google across the board for traditional search, leaving Google to deal with plane tickets and local search? ; - )
> All those things you hate about modern Google are necessary survival strategies for this model.
The reason I doubt this is because right now, I can find what I'm looking for through keyword searches on search engines that allow that, where I very often can't through the likes of Google.
Early Google was somewhat intentionally optimized for long-tail searches, the thought being that common searches are served fine by most search engines and it's the difficult long-tail searches where people actually try other search engines and cause people to switch their defaults/habits.
I suspect it's very difficult to properly define a machine learning objective to specifically optimize for the long-tail. Without very careful crafting, I suspect learn-to-rank optimizes more evenly, resulting in poorer ranking performance in the difficult long-tail.
Back 20 years ago when I worked on web search infra at Google, a bunch of the "do what I mean, not what I say" features were disabled if you used any of the search operators in your query. If I recall correctly, it even used its "Kansas" database of per-user data to check if you'd recently used any search operators. Most of that would be very difficult and/or costly to implement with learn-to-rank, though, I guess you could pass a "power_user" boolean to learn-to-rank to try and regain some of the lost functionality.
In fact, as of November 1997, only one of the top four commercial search engines finds itself (returns its own search page in response to its name in the top ten results). [0]
Circa 2003, Yahoo wasn't terrible. By our internal metrics, Yahoo search had better results than Bing before Yahoo started using Bing as their back-end. By that point, I suspect all of the major search engines were using some form of link-graph based ranking signals, likely modified both for scalability and avoiding the PageRank patent.
Maybe I was a bit loose with my definition of "Early Google", but in the 2003-2010 time frame, the major search engines had pretty well figured out the most common 95% to 99% of queries.
Most of the time when I have a tool that tries to guess what I want, I hate it. I hate it with a deep and burning hatred. I call it the "Clippy" effect.
Suggestions/autocomplete are force multipliers that can also go <1. When they are accurate, they save you a lot of typing/work. When they are inaccurate, you waste time processing the noise and have to do the work anyways.
Here's the problem (and why I personally hate overly smart IDEs) - my job is a creative one where most of the time I'm thinking and want my hands to be on autopilot. Writing out a skeleton class definition is something I can do blindfolded it takes almost no effort - an autocompleter getting that skeleton wrong (or writing a skeleton for a class with no interfaces and requiring me to backtrack through its garbled crap) takes effort. Reacting to things (observing the world and acting in response) will always take more mental effort than mindlessly slamming out some rote BS.
IDEs optimize a part of my job that I honestly could care less about (writing massive amounts of green field code) at the expense of making occasional mistakes when I'm doing my actual job (examining and tweaking existing code). This trade off is not worth it.
I'll uh - I'll keep function autocompletion though. Trying to remember whether my coworker called it `getWidget()` or `WidgetRegistry::get()` is something I don't need to waste brain cells on.
The history of useless code-automation tools is long. Back in the day, it was generating boilerplate API code for Windows etc. Write an app in a day!
But the lie was, in an app development schedule of 2000 hours, the tool made the 6-hour window definition effort into a 1-hour effort.
So what? It was good for demo-ware showing what an app might look like. But pointless for real work. Popular among the vaporware crowd, marketing and pitching etc. But not for software engineers.
CoPilot is interesting and I'm excited to see what happens with it but... IMO, if CoPilot is that valuable to us I think we've actually just failed (as an industry) to build a sufficiently high level language.
In my day job I produce a lot of infrastructural/lib tools for other developers to make use of and good syntax sugaring is worth it's weight in gold. Being able to write concise and powerful expressions without rote overhead can drastically increase dev productivity while reducing ongoing maintenance cost. If you've got a noise ratio near 50% you could probably benefit from some refactoring to kill a big chunk of whatever you're repeating so often.
Ah - and that's where it's really useful to have a set of eyes in the company watching for things your infrastructure needs to make everyone else more productive. The business specific rote stuff is usually a good idea to abstract away as well since it's so often highly critical and needs to be replicated often across the codebase and it absolutely must be consistent.
We a right at the point in history were this is about to change. I tried "new Bing" and it understands better than ChatGPT while also being able to use online search. And Google's Bard is just a few months, or weeks, away. Just a few years ago (pre GPT-3) this would have been unthinkable.
Google PageRank was a breakthrough, but IIRC, another huge search improvement was Google AND'ing search terms instead of OR'ing them: if you searched for apple pie, AltaVista would return a list of pages that include apple OR pie (perhaps a lot of useless hits about Apple Computer and cherry pies), while would Google return only pages with apple AND pie.
Much as I'm a bit of a sceptic, this is actually quite a compelling aspect of ChatGPT isn't it? It's not even the LLM necessarily, it's the interaction and the NLP: 'those results are not what I'm looking for [remove all those], I remember something I saw before [if you have access to my history..] which included the phrase 'hunter2' [refine for exactly that, not hunter without number, not different numbers, not synonyms for hunter]'
When I'm looking for web pages, though, I actively don't want to use natural language. Natural language is too vague and open to interpretation. To make it work, you have to get very precise in your wording, and it's just easier to get the right precision using keywords and operators.
Natural Language is great for loosely defining what you want, whilst literal searches are great for finding the exact occurrence of something you know you are looking for.
I can imagine something like:
Search for the error code: "XY123", triggered by product "ABC" when I try and start it bound to port 123. I'm using version 1.0.0, and want results from the official documentation or sites of good repute. Only include solved problems, and give me references in your results.
Granted, a lot of that could be covered by special incantations in a more literal search, but that just feels far more flexible.
Note I'm trying to delineate between literal (in quotes) and Natural Language in this example if it's not clear.
I suppose it depends on what you're searching for? In your example, you're looking for an answer to a very specific question. That's more of an "answer engine" sort of thing than a web search.
When I'm searching the web, I'm very rarely doing that sort of search. I'm more usually looking for a variety of websites that talk about a thing, not for a specific answer.
Is quoting terms not working for you? I don't think I've had problems with any specific searches when I explicitly quote my terms. e.g. "this is exactly the phrase I want to search" or "search" "these" "words"
You also have to go into the "Tools" menu at the top of the results and change "All results" to "Verbatim", otherwise it'll probably include conjugations even with the quotes.
Most likely, this is what you think you want. If Google actually gave you this you'd probably be wondering why your results were so irrelevant.
It's similar to how music players don't use true random algorithms for shuffling music. The truth is that what people want from "random" is "I am not so likely to hear the same song too quickly" not "true random," which would have the possibility of playing the same song twice in a row.
Your reasons for search usefulness declining somehow ignore the elephant in the room: SEO. Because search as a business tends to be about advertising monetization, that's the system that gets gamed.
Want better search? Or at least a differently-skewed search? Find a way to make it not be about ad money.
Recent example. I've got a pine64 a64 board with an ES9023 audio "POT" daughter board that runs and boots linux using armbian fine. Doesn't know about the ES9023 audio DAC, spdif and i2s(?) rca output. Ok so first thing I think of is, what kernel module is the driver for it so I know if it is even available or do I need to compile a kernel.
Sure pine64 documentation is woeful unless you want to read schematics or something so it could have been there, but it isn't. This should not be a hard question to answer with a search engine.
pine64 a64 audio es9023 linux kernel driver
Try it. Watch google do the wrong thing no matter what you do. Quote. allintext: you name it. It's nuts. Google wants to show you what it wants, not what you want. That appears to be what it is. You want to find something, get stuffed. Google is no use to find things. Google is good to show you things it chooses in the general area of your interest. Ok but that is not a "search" engine. It's a recommendation system, fine if that is what you want but a flipping diaster of orwellian proportions if that recommendation engine is the best we have for search and we haven't noted it and responded.
I'm more likely to get the answer faster posting about how garbage google search his than repeatedly trying different google "tricks" just crazy.
So yeah the answer I really want is what else can we use that actually does search?
So the question I want to answer is "What is the name of the kernel module that drives the es9023 daughter board on the pine64 a64"
Which of those results are you recommending as the answer?
edit: completely and totally prepared to accept I'm being a moron. The name of the driver, the link where you got it and how you got the link from google and I will decry it loudly here.
You're not being stupid, I couldn't find that either.
Although I do notice you wrote 9203 instead of 9023 in your suggested search. I get slightly more 9023-specific results with 9023 there: https://i.imgur.com/V5ty2In.png
I do think maybe you're expecting a little too much of any search engine for such an obscure query.
That's not obscure imho. What kernel driver do I use for this hardware is /exactly/ the kind of search google built its reputation on back in the old days when they did search.
The shock for me is I really /expect/ that to be easy becuase so very many similar searches have been, using google search, before. I've noticed it getting steadily worse along these lines in ways it really did not used to be. I've noticed it being because it wants to recommend not search. Seems others are noticing the same.
What alternative strategies do we have for search where we used to rely on google and now cannot?
I think what OP means is that there may not be a webpage answering this in the first place and thus there’s nothing for Google to index and return. Now of course, a sufficiently smart AI could know how to answer your query directly even if there’s no answer to your question made online. I don’t think the current batch of AI technologies can do that reliably.
Edit: also doing a search for “kernel module es9023” yields enough hints that to me indicates that there isn’t support for it even without it being a daughter board and thus there is no module. So the reason you’re not finding an answer is because there isn’t support for it and there’s no results confirming this 100% because it’s a very esoteric niche question. That you think it’s not is perhaps an inability to estimate how many people might be trying to run audio on this combination of hardware.
the hardware is NOT seen by the kernel. Does the kernel have the driver at all or do I need to compile a new kernel?
Ok so what is the name of the module that has the driver?
I'm sure that it ain't that hard to find out. I'm sure google with some google-fu used to give you an answer to exactly this kind of question in no more than 5 minutes of googling. 10 at most. I'm sure the driver exists.
Try it. Really. Watch google repeatedly drop and change your keywords, claim it hasn't then return pages with crucial keywords absent.
It's recommending. It's not searching. There is a difference and for this kind of thing it sucks very hard. You must have examples you've seen. This isn't my first it just came up in the last week so I picked on it here.
What makes you think there is any support for this hardware? By all accounts reports on the internet indicate that there isn’t any support so you’re asking Google for evidence of existence for a thing that doesn’t. And the topic is nice enough that it can’t find clear examples of counterfactual. This is also a bad use of Google. For something like this why not ask on the LKML? Something like this is perfect for that.
Bing Chat suggests the "snd_soc_pcm5102a" and "snd_soc_hifiberry_dac" drivers in creative mode, "snd_soc_es9023" in balanced, and failed to return results in precise. I doubt that's accurate.
I tried using Startpage and DDG to see if I could do any better. I mostly found people complaining that their ES9023 POT wasn't working.
PINE64's forum search returned a few topics with half-answers. Apparently a distro called "Volumio" works? There's a report of someone having success after modifying asound.conf on Debian Stretch.
This leads me to believe that there's not much useful information about the POT, or that it's being posted somewhere that doesn't get indexed.
An "answer machine" such as Ask Jeeves is a fantastic and desirable tool. So is a search engine.
I think a lot of people (with Google's encouragement) don't see them as two different sorts of tasks, but they are. In an ideal world, we'd have both.
If Google (or whoever) wants to be an answer machine, that's cool. My fear is that it will be done instead of, rather than along with, a search engine.
I'm surprised The Atlantic didn't mention where the Jeeves persona came from, they usually like that sort of exposition and they were already interviewing one of the creators.
Jeeves is a recurring character in a series of comic short stories by P.G. Wodehouse. He is a valet for the hapless gentleman who narrates the stories, and he has a miraculous gift for keeping his boorish charge out of trouble.
I wonder if chatbots will end up serving a similar role. Sitting on our shoulder and demurely asking whether the esteemed customer is quite certain that pizza should be ordered for dinner again.
For me, the killer feature of search has always been "never". Never show me a result from that source or any like it again. Never show me Pinterest, Ebay, or any of the thousands of Stack Overflow scrapers. I still search for websites, art resources, technical specifications, model numbers, book excerpts, and many of the other things the internet offered in the before times. Let me use filters and negative operators, as well as domain blocklists.
Maybe someone can bring him back with a pre-prompted ChatGPT.
me> For the remainder of this conversation, pretend to be the butler named Jeeves from Ask Jeeves.
cgpt> Very well, sir/madam. As you wish, I shall assume the persona of Jeeves from Ask Jeeves for the remainder of our conversation. How may I assist you today?
me> What is the capital of Assyria?
cgpt> I apologize, sir/madam, but Assyria is an ancient civilization that existed in the Mesopotamian region of the world, and it does not have a modern-day capital. The ancient city of Assur was the original capital of Assyria, but it has long since ceased to exist as a functioning city. If you have any further questions about ancient civilizations, I would be happy to assist you to the best of my ability.
A few years ago I was looking up US trademarks and randomly tried Ask Jeeves and noticed all the registrations were dead. I've just looked again and some random guy in Seattle has a registration going through for it as we speak.. so maybe Jeeves will return in some form or another.
As I recall, Altavista and Lycos tended to prove the most useful for me in the pre-Google era. Yahoo was hit or miss and Excite was middling.
I’d bet that the things being searched for influenced which engines were best for a given individual. It’s too bad there’s no way to see what each excelled at.
Yahoo's directory was created by humans, too, afair, which might be one solution to the current problem. It costs a lot to create and maintain, but in a sea of gamed websites, it might be a viable solution to the problem of finding genuinely good information.
Uhh we are talking about ask.com / iac and the same idiots who made the ask tollbar; Hey they also claimed they did yoga in their data center too . Let’s talk about that
https://archive.ph/xlOM3