I was searching certain words and the interface drastically changes based on what they think of the word. Then again, the auto completion is apparent hack of some gpt models in a rush.
It really used to be cool and stable, at least for me. Deep learning is not a magic hammer.
For examlle search for USB, Google will give me result of USA and then ask me back if I need infomation about USB.
And it's frequently ditching keywords from the prompt.
Sometimes I want to look for specific comments on some product for instance, I try different vocabularies that people might using exactly in the comment. But then all of these returns the same set of results repeated for 20 pages.
What is the point of Google search engine? Kind of trash to the point that it is now phenomenal.
Last thing is I noticed that image search is not even working recently. The average result is like 100 photos and done. Instead of infinite images in the old days.
I will frequently search for z foo bar z, get documents that have synonyms of foo and bar (and the synonyms will be highlighted in the snippet), then I will search "foo" "bar" and get the same documents except now it doesn't highlight the synonyms in the snippet. Maybe I am in some awful cohort where this feature is intentionally broken to see if it reduces engagement.
They used to have a bug with those
Same here, but seems to be 30/70. 30% of the time it works.
A while ago I was looking at an odd NetBSD issue and searched using "NetBSD". I got a few results on top, but most of the results on the first page were OpenBSD without any mention of NetBSD.
Once you get your query dialed in, prepare to take the Turing test.
Google Search today is only usable through frontends like SearX, and even then the results are no better than any competitor's.
Privacy concerns aside, ha ha, I wish a search site could rank results based on how long the user spends on the site. Most of the crap-sites that seem to get on the front page would disappear instantly from the rankings.
If I am searching for some particular piece of information, the best site is the one that has that up front, which means I get in and leave pretty fast.
Assuming that a user spending more time on a site is a good thing leads to large amounts of fluff (like with recipes...). But then again, there are some sites where that metric would be helpful.
prompt: Hey ChatGPT! Here is a list of websites. Make a subset of it that looks like they are on the 4th page in the search result!
It's gotten very frustrating to use because it tends to drown out good results with generic results from sites doing SEO optimization. There are some tricks around this. E.g. prefix any query related to AWS apis with stackoverflow or include some package names, use some quotes, etc. If you don't do that, you just get a 100 "this is why we believe AWS is so awesome" marketing to filter through. There are ways to mitigate that but they are quite tedious.
Bing on the other hand has rolled out gpt 4 based search. I've been using it and it's been pretty amazingly good taking over when Google fails me. Which it does frequently.
I particularly like it that I can ask follow up questions. It mixes answering questions and backs up the answers with links. I've been relying on it quite a bit since I got access a week ago. Very annoying that they make you use Edge for that but worth the detour from Firefox when Google is obviously not up to the job. Google is rightly worried about this.
Now the + and - tokens are just a mild suggestion. Wrapping a word or two in quotes sometimes helps, but more often than not just returns no results.
It's funny, in middle school we had a whole unit on constructing search engine queries. It was really quite powerful back then. But I guess it's less profitable for you to actually find what you want without spending 20 minutes sifting through ads
Lightest laptop? Here's are 10 lists of laptops that exist in 2023, with zero effort put into distinguishing them. As if SEO spam wasn't bad enough on its own, Google started ignoring your query to serve more of it.
Then there are specific queries that are reinterpreted as generic ones. You get 5 results or so, then it starts dropping keywords on a 2-3 keyword query.
It's infuriating.
To me, GPT-4 is a breath of fresh air that directly finds what I am looking for and answers complex questions quite well even if I don’t entirely know how to ask them perfectly. People like to talk about prompt engineering to optimize GPT, but I find I can get good results even with very lazy, sloppy input.
On Google, I often have to mess around with quoting certain phrases, using -foo to remove certain types of results, etc. Google is more finicky; you have to adapt to it, because it doesn’t adapt to you. Actually, in many cases Google attempts to adapt to you, but does it poorly, which is worse. Google changes results by location, even if I deny it access to my exact location, and shows barely relevant results from an imprecise location instead of location agnostic results. Google also changes the results based on search history yet it doesn’t really understand how one query is related to the previous query, the way GPT does. There’s also no “New chat” equivalent on Google to start from a fresh slate, short of clearing your browser data.
On top of all that, GPT has a paid tier to better align our incentives and avoid ads. Even the free version has no ads for now. It is a better interface by far.
The fact that I have to log in to GPT is probably the most annoying thing about it at the moment. It does also hallucinate sometimes and it’s too much of a prude. There are concerns about its power and bias, etc. But it is clearly a step-change in technology. One which seems destined to improve, not get worse, as Google has.
Kagi is priced for sustainability and has features like lenses (e.g., search your own list of niche community sites) and prioritization preferences (raise arstechnica, lower buzzfeed) deeply helpful for technologist searching.
Friends don't let friends Google Bing on DuckDuckGo.
Google is a lot better at keeping a profile on you whether you like it or not.
This is a thread about search quality, and neither of your objections lower search quality.
Further, neither of the objections makes any difference in the practical world where Google ties the search terms to your account, and you don't have to log in to search, in the same sense that you don't log in to Google to have your search terms tied to you. After using kagi, then you just use kagi, and they don't save your terms.
Finally, Google has assertions in the TOS you agreed to about what they can do with your searches. So does Kagi. If you do care about search terms tied to you, you're welcome to read those and understand who has concern or liability for what aspects of your privacy.
Put your money where your mouth is!
$5/m for only 200 searches (reminder, 1 page = 1 search. So if you hit "page 2" you just burned yet another search from your quota) is insane. [1]
Love it or hate it but Neeva is also on the private search engine market, and you can get unlimited searches for 50€/y. I find their pricing way more reasonable.
---
Search has a cost and your searches have historically been paid by advertisers. Recently, also by VC money.
With Kagi it is neither, nobody is paying for your searches but you. Price also should not be the only vector of comparison - quality of results (which we prioritize and are known for) speed and privacy policy should be considered too.
Of course I understand if none of that matters if one’s budge for search is $5/mo. In that case it will have to be VC subsidized search as it simply is not economically viable or sustainable to offer unlimited search (at least of Kagi quality) at that price point.
That isn't true at all from my experience. I just tested it and I was able to search for adult videos, etc.
I decided to try Kagi two weeks ago after hearing about it and trying their Orion browser on iOS. So far it has exceeded my expectations. Results have been on par with G/B/D. The new summerizer has been pretty good and I enjoy getting summary of the results at times to save me time (although I've had a few it couldn't summerize or wrong results in past two weeks). I also love the ability to prioritize sites for your future results and the summary ability of individual pages.
I will stick with the $10/mo for near future. It is small price to pay to support their current goals and intentions.
But I'm a little confused by the exact wording of this post. Google's Bard is in competition with GPT, isn't it? Is it not? Was there a news item I missed about this?
I don't use Google often enough to have a strong sense of what might be different and what remains the same, but I've just run a couple of searches and don't notice any drastic interface changes?
This is not a complaint -- I just really want to know the new reasons to complain about Google!
I'll Google for something and see the results are not very good, but I'll see in "People also ask" a question that is asking about what I'm looking for, in more detail and better expressed than what I asked.
It used to be that clicking that would almost always reveal what I was looking for.
Now it usually just gives me a result along the lines of all the bad results to my own search.
I don't think Google ever had a great algorithm and switched to something worse. It's just the SEO actors putting crap out there now, and clickbait. The thing it became worse at was recommending small sites/blogs over big publications/sites.
Back in the day, all they had to face was people trying to spread viruses but those results were super obvious. And they weren't even flagged back then, just users were smart enough to instantly spot it from the result.
Spam sites generally try to either show you lots of ads, make you click on affiliate links or outright sell you something.
This can all be detected and used as a negative ranking signal, so that all else being equal, a page without advertising/affiliate links/etc would rank higher than a page with those elements. This would allow non-spam content (if it exists and matches the query) to outrank spam content.
Given Bing and DuckDuckGo don't have the same incentives I wonder if their results are any better.
Microsoft has turned their entire ecosystem (including OS) into an advertising-saturated cesspool.
DDG might be "okay" for a while just because they're still trying to gain marketshare, but the underlying incentives are the same.
A paid search engine is the only solution that aligns the end-user's incentives with the search engine provider's.
Google themselves don’t even know how their search works anymore ?
When I first started the low hanging fruit was these so-called long tail keywords. long, low traffic search terms that represent a specific question within a broader topic.
"how to invert a binary tree with recursion in python" would be a long-tail of "how to invert a binary tree".
A fairly small blog could rank for these lower-volume keywords without much work or investment. during this time, a low quality spam blog would have a large collection of barely-releated articles targeting very specific searches. Their site structure would be something like:
computerlinuxhow2guide.net: - how to invert a binary tree in python - how to quit vim on arch linux - how to make an open world free-to-play video game in fortran - how to install docker on ubuntu 18.04 - how to build a flappy bird clone in react native for samsung phones
Webmasters looking to build a more sustainable business would snipe a few longtails when possible, but their overall content would be topical and more cohesive.
datastructureslut.com: - queues - build a queue using an array - build a queue using a linked list - binary trees - when to use a binary tree - how to invert a binary tree etc..
Google started to pages from sites with a lot of topical relevancy. Covering all the most popular search terms and interlinking these pages would give you a big boost. Affiliate bloggers responded by creating sites like "fishtankexpert.com" and "lawnmowerguru.com".
There are a LOT of these sites. Case in point: the examples above are just the first two things I thought of for $product $authority. and they both exist.
It became more common for larger sites to incorporate longtails into existing articles. Instead of writing a separate article, "how to invert a binary tree in python" could just be a subheading in a larger article. Tools like surferSEO became popular, before long it was common to see 10,000 word articles that covered thousands of search terms.
6 years ago, it'd be pretty common to find a forum about fishkeeping or a personal blog about landscaping on the first page of the search results for these kinds of keywords. But even then a lot of these sites would be abandoned or out of date. That made them an easy target for new webmasters to outrank, even with a limited budget.
The last major factor is profitability. With top 10 style articles, broad searches such as "best fish tanks" would bring in less sophisticated users. We'd often see purchases come in immediately after clicking on an ad. Most sales were one of the top 3 items on our list. Most visitors would skim the headlines, or read the top 1/3rd of the article.
specific Terms like "55 gallon saltwater fishtank" would bring in users who spent more time on your site, were more likely to read multiple articles, and more likely to return to our site at a later date. Overall, longtail traffic was more likely to make a purchase, and they would spend more money. Those sales often came in a day or two later, while broad traffic tended to purchase immediately.
Google never seemed to like longtail traffic. Not only because their pagerank algorithm was easier to manipulate, but also because they were less profitable for ad sales.
When customers could bid on highly targeted search terms that would reduce the number of people bidding on terms with low commercial value. In fact, if you set up an adwords campaign today you won't even be able to see traffic stats for these terms. They are obfuscated until you meet a certain spending threshold.
If you do optimize your bidding to avoid irrelevant searches, you'll get big warnings on your dashboard about "errors" they can "fix" for you so you don't "miss out" on all this "great traffic".
Since niche specific forums or blogs were basically dead, and these terms were almost entirely dominated by SEO spam, and larger "trusted" sites had started covering longtails in their broad content, google started to group topics and redirect searches to larger, more popular results.
The upside is that spam articles like "how to change the tire on a 1994 corolla" wouldn't easily outrank an article that has more general information about tire changing. There will be many popular articles targeting "how to change a tire", and "1994 corolla" is probably not mentioned on any of them. So google will assume that the uncommon portion is irrelevant.
When you have an error code, this strategy falls apart. The most common keywords in that search will be used in many different errors, while the portion specific to your error may only have a few results.
While there is still a lot of spam, these are not the low effort traffic grabs from a few years ago. These are sites that have invested heavily in content, back-links, and behavioral optimization. they'll be acceptable to a good chunk of the people that visit, and their advertisers will be more closely aligned with the results.
who needs search results anyway. If your question can't be answered by an advertiser or the knowledge graph (https://www.google.com/search/howsearchworks/features/) then it's probably not worth knowing.
google.##.g:has(a[href=".pinterest."]) google.##a[href=".pinterest."]:nth-ancestor(1)
What's the good search engine these days?
Let alone how are they affording to index and crawl all those pages, which is still only a tiny fraction of the entire internet.
Maybe search engines are a dying breed, and we'll have to rely on AI and word of mouth for information soon.
Which apparently worked well for quite a long time, but also dangerous over time because you don't really know why something now ranks well, or doesn't. The best you can do is have human raters score various searches, tweak a bunch of weights/knobs, and try over and over again.
Meaning, perhaps it's not that the internet is too big, but that the overall model is too big and disjointed to manage well. Also, if you're on this team, you're competing with the ad side that's continually pushing the organic results down the page, where they matter less. A lot of fussy tuning around "quaint hotel in Paris" seems wasted when the end user is 3 scroll events away from the organic results.