The trick is that Google actually works great. It works better than before. It just doesn't work in the way that we (highly educated and technical people) want. We all laughed at our parents and grandparents for typing full sentence questions into the search bar in 2005 instead of carefully crafted incantations with keywords, "site:" labels and other such things. Young people often use full sentences today because that is what works.
The google we grew up with was a tool that allowed you to precisely retrieve authoritative writings related to a subject, but today's google is a lot more like ChatGPT than that.
No, Google results are trash. Here are three examples from this week:
- I Googled "speed work running", wanting some workout suggestions. I visited four or five results that either didn't have concrete suggestions, were poorly written, or overly verbose. I ended my search with little faith that I was getting good advice. I typed the same thing into ChatGPT and got a suggestion of 6 different workouts that all seemed coherent.
- I asked Google Home "How do I make overnight oats". It replied, "I found a result for [...], should I read it?" then "The first ingredient is Nutella". That's it, that's all it said. I tried the same search on the web, and every result Google result was spam that was 20 paragraphs of inane blog content with a recipe tacked on the end. Again, ChatGPT then gave me a sane, baseline, recipe.
- I searched Google for "how much caffeine is in coffee". It gave me a calculator that said, "40 mg" and then a suggestion an alternative search for "Q: How much caffeine is in an average cup of coffee? A: between 80 to 100 milligrams". It turns out the calculator was normalizing the caffeine content to caffeine per 100mg of coffee.
I'm using Google exactly how their product is training me to use it, but the failure modes are all consistent. Google's AI features has not real understanding of the world, so their instant answers are frequently nonsensical. And Google clearly can't filter out SEO garbage.
Even if they could filter out SEO garbage, Google's early success killed the golden goose. 99% of web publishers are publishing content for Google, not for their readers and web publishing has become a cynically commercial affair. Individual publishers have by and large moved on from the web to other creator-focused platforms. So, the quality and experience of web content is absolute rubbish. Results are filled with cookie banners, ads, signup prompts, verbose SEO filler, poor writing, lack of authority, etc.
You are using it right however your bias is evident on the results. A 20-something will just scroll through the blog content to get to the recipe at the end. No huss, no fuss. In fact, they’ll open multiple tabs of search results to compare recipes in the time it takes you to gawk at the fact that you have to scroll to the end. Where is the recipe? Why is there a video? Table of contents? No, a 20-something will just flick their screen wildly until they are at the bottom, and scroll up. Give me the results and backtrack to where I’m at. This I think is a result of their upbringing and everyone’s focus on results rather than process. So you may believe you are doing everything correct but you are just going through the motions. The web today is different than it was even 5 years ago.
I wish I could search like I used to but we are where we are.
> A 20-something will just scroll through the blog content to get to the recipe at the end. No huss, no fuss. In fact, they’ll open multiple tabs of search results to compare recipes in the time it takes you to gawk at the fact that you have to scroll to the end.
Uh, I do that and I’m 37. But I don’t scroll, I click the button that says "recipe" that’s on pretty much every recipe site.
IMO this is mainly about being used to the state of recipe sites on the internet, and has nothing to do with searching.
The state of recipe sites is 100% driven by Google. Those sites add the fluff because that is what Google rewards. The actual recipe quality doesn't matter. If I use my example, the page contains:
* Bio of the author - Google SEO rewards content written by "real humans", so this is there to appease the bot.
* Giant images of overnight oats - This helps the page rank highly on Google Image Search.
* Fluffy paragraphs - "The beauty of overnight oats is that you can make them as simple or creative as you’d like. The base recipe is delicious, and filling all on its own. But if you’d like to spruce it up, you can add a variety of toppings and mix-ins, including fresh fruit, nuts, seeds, spices, jams, and more". Is that supposed to be text written for other humans? Google rewards original content, which means recipe sites need to include massive paragraphs of filler to appear like an authoritative source.
* Ads - Gotta make the money.
* More fluff "Here’s a few reasons why you should to whip up this recipe today…" - Again, this is not text written for humans.
* Repeat the fluff, pictures and ads approximately 5 times.
* Common questions "Do you have to use yogurt?" - Trying to appeal to Google Instant Answer searches.
* Recipe - Finally!
* Recipe rating - This allows the site to put a star rating and review count on Google, again, for SEO
So, 95% of the content and functionality of that page is an attempt to convince Google into sending traffic, or ads generating revenue, rather than serving the end user.
Content on the web has dramatically changed from 20 years ago because publishers are creating content for a broken machine and not for other humans. This negative feedback loop is hurting the web, and not just Google. This leaves the door open for upstarts like ChatGPT or TikTok to gain mindshare.
So the fluff is both an attempt at that substantial literary expression, and a way to differentiate this version of the recipe from 1000 other versions of the _exact_ recipe because the recipe itself is not unqiue or copyrightable! Without it, recipe's would be penalized by the duplicate content penalty Google applies.
I think this is more about ads then copyright. All these recipe blogs I've seen still follow with an uncopyrightable list of ingredients and simple directions. The recipe still isn't worthy of copyright, even if the content accompanying it is. Sure, the author may be working on something they are passionate about, but the site itself is all about the ad revenue.
Sorry, I might not have been precise enough, I did not mean to say that this is not because of Google, only that knowing how to use recipe sites is very distinct from knowing how to google, you even encounter those sites no matter what SE you use (I haven’t used Google in years).
The fact that young people have gotten used to inefficiencies and have gotten good at partially working around them does not mean the inefficiencies aren't there.
This often works the opposite generational direction: when a young person has a flat tire they often call a road service and wait two hours, while the old guy would install the spare himself in less than 8 minutes. But either way, flat tires are inefficiencies that are best eliminated.
Looks like what you're saying is that the results are garbage but people who don't know it could be better are fine with it, because they learned it's how the world works and can't be otherwise. While the ones who are cursed with the knowledge that it can be are left to complain.
I learned a trick recently. If you want "copyright-challenged" results, or exact results, try Yandex. It's a worse search engine and, therefore, better at matching exact text.
>A 20-something will just scroll through the blog content to get to the recipe at the end. No huss, no fuss. In fact, they’ll open multiple tabs of search results to compare recipes in the time it takes you to gawk at the fact that you have to scroll to the end. Where is the recipe? Why is there a video? Table of contents? No, a 20-something will just flick their screen wildly until they are at the bottom, and scroll up
I'm over 40 and do that (and the fuzzy stuff / open questions in the top comment).
In fact, most 20-somethings are much worse than this - some studies even pin them as worse with most computing use that is not about social media apps.
If that's your take, it says more about you. We are ALL victims, to something. Sometimes it's our own fault (IP4), sometimes it's not (adtech). Next time, if you feel like a comment is negative or gas lighting or victim blaming or cancel-culture-worthy, you should state your reasons why you feel this way. Instead of stating your opinion as fact. No matter how right you think you are, there are always other perspectives (Israel/Palestine).
While reading your Google examples, I don’t think I could parse out what you were intending as a human. Perhaps different keywords would work better?
For example, “speed work running”, were you looking to improve sprinting? (I’m not a runner, so it almost reads to me like “working fast while on a treadmill”). Reading the rest of your example, maybe “sprint workouts” would give better results?
(40+yr old guy here, Google had always worked for me, as many people had described it working for 20yr olds. Finding the right, specific enough set of keywords had always been my “trick”. Little did I realize I subconsciously generate embeddings like a LLM to Google search)
"Speed work" is a class of running drills. The term is very common in running. I'll let ChatGPT explain. This was its response was given the same prompt, when I specifically asked it to give me options it's answer was more specific:
Speed work in running is a type of training that focuses on improving your running speed and anaerobic fitness. It involves various structured workouts and drills designed to increase your stride turnover and overall pace. Speed work is essential for athletes who want to enhance their performance in races or simply become faster runners.
Common forms of speed work in running include:
Interval Training: Interval workouts involve alternating between periods of high-intensity running (fast) and rest or recovery jogging. For example, you might run fast for 1 minute and then jog or walk for 1-2 minutes to recover. This process is repeated multiple times.
Fartlek Training: Fartlek, which means "speed play" in Swedish, is an unstructured form of speed work. During a fartlek run, you mix bursts of faster running with slower paces or jogging. You can do this based on how you feel or choose landmarks as your guide.
Hill Repeats: Running uphill at a high intensity is an excellent way to build strength and speed. Hill repeat workouts involve running up a hill at a fast pace and then jogging or walking back down to recover. This is repeated for a set number of repetitions.
Tempo Runs: Tempo runs are sustained efforts at a "comfortably hard" pace, typically just below your anaerobic threshold. This helps improve your lactate threshold and race pace.
Track Workouts: These are structured speed sessions done on a running track. Common track workouts include 400-meter repeats, 800-meter repeats, and 1,600-meter repeats, where you run at a fast, consistent pace and take rest intervals between each repetition.
Strides: Strides are short, fast runs of about 100 meters that help improve running form and leg turnover. They are typically done at the end of an easy run.
It's essential to incorporate speed work into your training regimen gradually to avoid overuse injuries and adapt to the increased intensity. Make sure to warm up properly and cool down after each speed workout. Consult with a coach or experienced runner to create a personalized training plan that suits your goals and fitness level. Additionally, listen to your body and allow for sufficient recovery between speed sessions to prevent overtraining.
If you search "speed work running drills" on Google, you get relevant results. It kind of sounds like you are cherry picking the "bad prompts" you sent to Google and comparing them to good prompts sent to GPT-4.
They specifically said that they used the same prompt. If I feed the exact string "speed work running" into ChatGPT 3.5 I get back essentially the same results that OP includes.
Bad prompts for google can have exactly the same words as good prompts for ChatGPT. The original commenter probably uses both of them at least 10 times a day, so it's not unreasonable to assume that they will run into those examples.
I would typically assume that a three word prompt like that is better for Google than it is for ChatGPT. Google is supposed to be good at keyword search, and ChatGPT is designed for longer prompts.
That means that your prior assumptions about Google are wrong. Google does a substantial amount of "semantic understanding" before doing a search. It is no longer a keyword matching tool - that stopped being the case a long time ago.
Also, ChatGPT is usually really bad with long prompts when I use it, and is often kind of close when I give it a short prompt. I assume part of this effect is psychological though, where you expect the short prompt to have worse results.
Mid-30s here, everyone in the shop at work uses me for Google because I still know how to use it effectively. Speaking with lots of people in person and via chats/forums has gradually made me come to realize how refined my search abilities are, relatively speaking. I think it's a combination of factors, like you mentioned "finding the right, specific enough set of keywords" is intuitive for us, but a lot of people just type in questions like it's Ask Jeeves from decades ago. Another major factor is one of grammar and vocabulary, because Google will fuzz your search string. In my experience, its "thesaurus" will rarely add more technical or specific words, so including them in your search will often give you actual technical data from a given field. Of course, this only helps if you have the proper lexicon to begin with, which was my main use case for Reddit before it went to shit. Hobbyist or technical subreddits were great for drilling down and winnowing out the specific name of something I was trying to find more information about. Now it's just more SEO LLM-generated nonsense trying to sell me crap I don't need. The other "secret sauce" I use when Googling is search operators, but we actually learned about those in elementary school (before Google existed!).
FWIW, my human brain did not understand your first query, in the second you conflate voice search with Google, and re the last, if I try it, it shows me a dropdown to select what I meant…
(Generally just confused by how ppl always claim it’s trash or it got worse but there never is a solid benchmark)
I think the problem with Google is that 'garbage' is highly subjective, and what you think is awful is actually highly engaging for a lot of users. Google has massive amounts of data about what results users click on, how much time they spend on a website, how much scrolling they do, etc. To technical people we see that as SEO 'trapping' people to push engagement rankings up, but the reality is that it's just engagement. People could leave a site of they wanted to, but they don't. They scroll through those 20 paragraphs of recipe back story. That data means Google are finding accurate results.
If anything this shows that Google's results personalization isn't taking your engagement into account. A problem that will get worse as more people block analytics.
- I searched Google for "how much caffeine is in coffee". It gave me a calculator that said, "40 mg" ... It turns out the calculator was normalizing the caffeine content to caffeine per 100mg of coffee."
The problem with "SEO garbage" is that SEO is highly adaptive - whatever change Google does, SEO will almost immediately find a way to adapt and stay relevant. There is just too much money to be made for those involved.
No the problem is that Google did what they really, really promised never to do: to optimize for financial gain, keep people on the Google site, clicking on ads instead of telling them what they want to know. Like one of those 10-episode Netflix documentaries.
Precisely. The main problem with google is that it's extremely hard to align yourself with the goals of humans with human needs, so they have to look at proxy metrics to rank sites before anyone actually hits them. Their current proxy metric appears to be that there is a lot of human-sounding text on the page, and so people optimizing for that are going to put a lot of text on the page, factual accuracy be damned. In the naughts, their metric used to be that many sites linked to yours, and so people optimized that by creating webrings.
you can't copyright a recipe itself, so in order to copyright the content on a website with recipes, you have to do that "My mother used to make this on dreary winter days, and we'd sit by the fire ..."
Now, i was told this, repeatedly, over and over. It's probably false, but if it is false, cite the decision that precedented this.
If you call "working" showing me an endless stream of blogs of some dude, an automatically translated russian clone of stackoverflow, a completely unrelated stackoverflow question, above the actual official documentation that it should be showing me… sure then it works.
To me what this shows is that it's really painful to try and use an "intelligent" algorithm for "content mediation" as if it were an actual search engine. Anything that spectacularly fails the "exact substring match" use-case is well on the way to becoming a content-mediator, and potentially as bad a content-moderator/thought-police/etc.
Unfortunately even before anyone started talking about "semantic web" or whatever, substring-match was declining in utility due to spam, so a substring-matcher was only as useful as it's ranking algorithm. So some cleverness in ranking is necessary, but when ranking becomes too clever, or irrelevant due to the arms race with spam, it's an almost inevitable slide from necessary rankers to mediators.
Since all the search/spam arms race is pretty much played out with giants like twitter/google anyway at this point, it is kind of hard to understand why we can't get and keep a separate and truly basic internet search engine that just works. I don't have a handy example, but in my experience even DDG & friends seem to routinely fail the "exact substring match" test for unusual phrases that I know I read verbatim last week.
What I (and I think others) really want is just substring-match tool plus a very minimal amount of extra cleverness for basic stuff like spelling variations, some fuzzing for dates/digits, directly adjacent stuff according to a simple thesaurus.
Not my area of expertise, but I would guess that there's a combination of problems that prevent this tool from existing.
a) Other search companies are all trying to do fancy-but-fuzzy "semantic distance" searches to emulate Google
b) It's still true in 2023 that just keeping an up to date index is a super hard problem no matter how many crawlers are in your army
c) Despite bots everywhere, the deeper web in social-media is just too login-walled for the rule-abiding crawlers to deal with
d) There's just no money in do-what-i-say-not-what-i-mean search tools
DDG seems unusually susceptible to highly ranking AI-generated overly verbose garbage. I can't recreate the same issue with google or bing, so I am not sure where it pulls it from. I can find those same results in google, it's just they are on the 2nd or 3rd page.
I've never really found docs particularly useful. Usually, I just want a few simple examples to get the gist of how something works and docs tend to be very bad job of that. Meanwhile, the W3Schools SEO spam HN disdains tends to good job of that.
Eh, I've had the opposite experience. If you search for anything slightly obscure in a language like Java, you'll get dozens of spam sites aimed at beginners that answer a much more basic question than the one you asked. That's because Java searches cater to a large population of people who can barely code.
Java is particularly rough because there are several populations of Java users (students, Android developers, enterprise developers, among others) and their coding standards are universes apart.
If I wanted the docs, I would get the docs. The blogs are useful because the documentation is largely unhelpful due to being overly simplicity or extremely nondescript. The blogs are where it's at because you can see how someone else used that thing
That's too generalized and IME not even true for say, Python docs, or regexes, or Gnu tools, or HTML.
Most bootcamp/academy/coder-type blogs/copycats are aimed at giving only a tiny fraction of the information, but with examples (sometimes only very simple ones that cover only the most basic use-cases, and no links or mention of the other missing 90% of information (e.g.: "to read more about regexes, see ...").
"site:docs.python.org/3/ term" gets much better results. (You have to add the site: qualifier, due to SEO it's no longer just enough to say "Python docs". Hence
"If I wanted the docs, I would get the docs." is not as trivial to do as you suggest. Things actually went beyond annoying to downright dangerous when searches for "Python 3.10 <topic>" were drowned out by older e.g. (3.6-3.8) version SEO stuff, instead of the latest Python docs.)
Queries like "awk by example" or "egrep by example" or "jq by example" seem to hit the sweetspot between completeness and working example: they give you third-party non-commercial unaffiliated expert bloggers not stuffed with upsell links to a bootcamp. Just pure information sites.
Having mentioned all this, now I've probably inadvertently cursed those sites...
Totally agree! My point is that they both probably contain the same answer and I bet that's all most people care about. I think most of us here have strong preferences for picking the "most appropriate" source, but frankly, I suspect we're weird.
I think Google Search "works" in the same way local newspapers "worked": they provide easy access to what most people want to know in their daily life, including gossip, the weather, the popular events and discussion topics ("melting houses")
If you're a German living in Morocco trying to understand Quebec's immigration policies in the original text you'll be fighting the search engine at every turn.
This feels like a good take. I’ve recently been trying to find more information on caveats of Section 179 of the US tax code, and at almost every turn the consensus is to go to a professional(which is what I should do), but for some nagging off hand questions, I was hoping to find some meaningful discourse on it.
Exactly. Google search is not a knowledge finder for power users, it's a content finder for normal people. The fact that it was good at finding knowledge for 10 years or so is entirely incidental to the goals of modern Google.
Seems like we live in parallel worlds but I will show you how I am right except you can really proof I am wrong:
At Google, they recommend to go to https://www.google.com/advanced_search so you can write all the keywords you want. In this case I searched for: kvt reverse exploit [1] you can quickly see that google search results are inaccurate and it shows first my HN post because it was very recent. Also the recent article where it is also mentioned and added after I post it here in HN [2] and the following results doesn't end in some correct previous post. I then try again [3] with a more precise search adding kconsole and only found three results without all the keywords while the advanced search says explicitly that all these keywords should be there.
Quoting as ` kvt “reverse exploit” ` gets me only a few results on Google, period. That’s true even telling it to include ones omitted as redundant. They’re all either your HN comments or blogs mirroring them. The original doesn’t appear.
That tells me that the main problem may simply be that the BugTraq post on seclists.org isn’t indexed by Google at all.
While that’d be annoying, I think it’d more likely be a specific configuration or robots.txt issue with that site than being a general issue with how searches are performed.
FWIW, the same query in Kagi found the original as the second result, just under a blog mirror of your 10/20 comment and above the comment itself. Since Kagi sources from Bing (along with Google) results, that reinforces the theory that it’s an issue specific to Google’s crawl.
Contrary to your other replies, I do think your style of query continues to work well on Google (where indexed of course)—-and so do full sentences.
I honestly think the issues when there are problems finding things usually come more down to 25-plus years of searchable internet history accreting a lot of clutter (especially in spaces like tech where old info ages out but never gets deleted), along with cynical SEO from low value sites deliberately skewing results for as long as the site remains indexed.
Neither is a Google problem, and the recency bias you observed is arguably the best way to combat both. Ads and site promotion are another story, and the reason I’m on Kagi, but I don’t think that hits tech as hard as consumer spaces.
Yes, you're missing the point of the poster above you. He's saying that the kind of technical searches you are trying to do are irrelevant to 99.99% of internet users and Google is optimized to give superficial but relevant-looking results to those kinds of people. For that metric Google does "work well"
So it's not that Google doesn't work, it's that you're using the wrong tool for your use case. If you want highly technical content or precise keyword search you shouldn't use Google. It's like you're going on Tinder to find a marriage partner.
It's not yesterday, it was 20 years ago when Google excelled at exact keyword matches. It has been a constant evolution away from that.
I hate the new Google as much as others, but if you don't adapt your search habits for 20 years when the whole ecosystem around you has been obviously changing, that's kind of on you. Just use another search engine that fits your use case. Personally I use Kagi and I haven't touched Google for the last year at all.
seconding fastmail - i was grandfathered at $5 a year for years, i pay $15 now i think - and i am probably going to pay for kagi (or ilk). I run nextcloud, after trying other things like syncthing and owncloud - as well as owning a synology i would never put on the public internet.
I think perhaps the problem was that when you were growing up, the Internet had a much higher percentage of authoritative writings related to a subject, and today's Internet looks a lot more like ChatGPT-generated drivel. Google's results might has changed, but I suspect less so than the Internet as a whole. Older Internet content was weird, and often wrong, but it wasn't an endless expanse of nothingness camouflaged to resemble human content.
While this is true, Google itself is in large part responsible for the internet turning into this wasteland. Most of that garbage content is produced in the name of SEO.
Yeah, gone are the days when you could enter a range of 16-digit numbers that started with "4", filetype:xls, and it would return a bunch of Excel spreadsheets from businesses that were storing credit card numbers in plaintext cells.
Noticed this as well. I've gotten so frustrated with poor results for things I need, I've been using "Verbatim" mode to force Google to stop interpreting my terms. This was a pain, so ended up installing an extension that forces verbatim on every search.
Recently, I searched "(-)-BPAP" -- with the quotes -- looking for the chemical benzofuranylpropylaminopentane, which is typically referred to as (-)-BPAP. EVERY result was for a BiPAP ventilator.
So I tried verbatim. It still didn't work.
Verbatim + quotation marks. Still nothing.
I guess New Google Search simply doesn't recognize the "(-)-" part of the search term. But this is characteristic of its recent performance. I can't even count the number of times Google disregarded part of my search term and gave me an inane result.
> Google Search simply doesn't recognize the "(-)-" part of the search term.
AFAIK, Google search has always ignored parentheses and most punctuation symbols (other than ones that are special to it, like +require_term -exclude_term "...")
I’m sorry but are you trying to speak for all highly educated and technical people?
Because as far as I know, I’m both and I find Google just as good as ever, if not much better.
The only place I see people complaining about Google is Hacker News and certain parts of Reddit and by no means are the opinions on these sites anywhere near universally shared by most people.
And by all means, I think calling yourself highly educated but not being able to adapt slightly when a tool isn’t working right a little rich…
there's no reason a person can't be educated and inflexible. in fact the two often go hand in hand. also, I'm not sure you're using "by all means" correctly
Yes, I have SafeSearch turned OFF. This is happening to pretty much anything that's remotely NSFW on both google.com and amazon.com -- it's getting completely impossible to search for perfectly legal things that are NSFW.
>It works better than before. It just doesn't work in the way that we (highly educated and technical people) want
I don't think so. I've been using the "full sentence questions" and fuzzy questions since forever, and very seldom use "site:" and other such more formal constraints you mention.
Despite that Google results have been getting worse for the past 5 years at least.
Google does this so people will execute more queries, stay longer on the search results, and see more ads. It's classic monopoly exploitation of abusing a captured user for increased profit.
They're perfectly capable of making things better, it's just more profitable if they don't.
You use mostly ChatGPT for finding information? When I use it I still double check it's output with a google or duckduckgo search, if I don't something I know is outright wrong.
Indeed, I've had to force myself to start searching in full sentences much more as the results from the old method of keyword searching have steadily deteriorated.
I disagree, but it depends on your definition of "working well". Yes, Google works better for my mom these days if you measure whether she is happy with the results she gets. However, the results she gets are mostly misinformation or spam. She just doesn't realize that's the case and is happy with the results regardless.
Measuring whether or not results are relevant to the query is the wrong metric to use. You don't want highly relevant misinformation. You want good information.
The google we grew up with was a tool that allowed you to precisely retrieve authoritative writings related to a subject, but today's google is a lot more like ChatGPT than that.