About the "dead internet conspiracy" - I've worked in writing how-to articles for a fairly large "help" website. They paid very little attention to the quality of the articles. I was paid for each piece and thus had about 30 minutes to write an article and later integrate feedback from internal review. Otherwise the payment became too low.
The most important factor was cramming SEO terms and links to keep people on the website into the articles.
The result is trashy articles that could well have been written by a bot but aren't. This could possibly be done with the help of curated bot-content, but I think we're far away from the point where this is really more profitable than getting students to do the work.
>> This could possibly be done with the help of curated bot-content, but I think we're far away from the point where this is really more profitable than getting students to do the work.
It may be becoming borderline. I expect that sentence/paragraph completion is already becoming useful to people who churn out quick content for a living. In any case, the important part isn't whether or not it's bots. The important part is whether or not it's authentic. The precise meaning of authenticity gets squishy, but it exists nonetheless.
IMO the sentiments are correct, whatever the details. Part of why google sucks is that the internet is worse, for a bunch of the things we use google to search for. The internet becoming a larger, more profitable industry changed it. Instagramming for influencer perks, SEOing, or selling targeted ads like FB do... it does not lead to the same places that earlier iterations of the WWW produced. Times change.
My friend briefly had a copywriting job writing weed strain descriptions for dispensaries. He was never provided the product he was describing, just told to make it up.
I have the very same feeling concerning electronics. Searching for a particular product does not even popping up 5-10 comparison articles but the content of all seems to be based on technical specifications of the manufacturer only, which I already have a hands on.
Significantly more time required for consciously choosing a product to purchase (which in my case is critical because I am like Sheldon Cooper trying to choose between PS4 and XBOX One normally, to the horror of my wife, she wants a new TV and it is months long project based on accurate and quick data, and now this, with Google, which makes our family atmosphere even more tense : ) )
In case you haven't encountered it yet: https://www.rtings.com is a good site for TV specs/reviews specifically. I know someone who works there and their methodology seems legit (I like it more than Wirecutter).
Most reviews are useless from the start because most reviewers are totally dependent on manufacturers or dealers providing samples (yet will generally claim to be "totally independent"). That's before you get to reviewers who can't or don't know how to test the product in question and so end up narrating the manufacturers specs to their faux testing.
Niche, probably outdated, but indicative: The only way to purchase a laser printer with high printing quality these days is to buy something expensive and hope for the best. Magazines used to do actual reviews of these things.
Brother MFC laser printers have delivered for me for over 15 years now. I have bought many for various small offices and my home and relatives’ homes, etc, and I have not heard any complaints.
I especially like the scan function where brother web connect OCR’s the document and saves it as a pdf directly in your Box/Dropbox/OneDrive/GoogleDrive folder. Just wish it worked with iCloud Drive.
Oops, my bad, I did not properly take into account high quality printing. It is definitely too cheap to be anything high quality, they just get the job done for everyday printing at a low price.
I look back at the newspaper stories I wrote a few decades back. I could get the score from the coach, find out hits from who and when, and after that 20 second interaction, I could write a news story in maybe three minutes, which told you everything you needed to know.
Somebody who worked at a winery once told me that the flavors they mention on the bottle are actually what the wine is missing, and they name them in the hope that the power of suggestion give you a more balanced impression.
It would be interesting to see some sort of study, to see the impact on wine labels on: normal people, vs wine "experts", vs sommeliers.
I've seen a few of those wine documentaries about sommeliers, and they certainly made it seem like it was a legitimate ability to identify stuff. I'd be interested to see how close they are in a more neutral, measured environment.
>Tasters can't even tell reds from whites with a great deal of accuracy. [2]
Actually, this one is bullshit. While it may be true that the average person without blind tasting experience cannot do this, anyone who has actually done blind tasting seriously should be pretty accurate at this. Purely structurally, red wines generally have much more tannin and more alcohol than whites.
Sommeliers are professionalized, with courses & exams. But I don't believe they are actually judged on their ability to taste wines and detect flavours. Designing such tests would be simple - rate of successful flavour identification based on data from other Somemeliers.
Instead Sommelier exams are subjective - candidates are judged by another Sommelier on subjective criteria rather than objective measurement. In my opinion they judge it as a dramatic performance: how quickly can the candidate rattle-off various flavours? How "high class" is the language they use to describe the flavours? How does the candidate present themselves? Sommeliers dress in fine suits, but this should have no impact on actual wine tasting ability. Yet I am sure if I showed up in a draggy old t-shirt I would fail regardless.
The whole industry seems allergic to objective scientific measurement. I am sure Sommelier's do have some ability to identify varieties, but are they actually tasting all those subtle hints they list off? I doubt it.
What is this comment based on? To me it feels like this is what you imagine the exams are like. The MS exam (the highest and most famous one) has three parts: theory, service, and tasting. The service part is all about presentation, absolutely (that's the whole point). The tasting part is not. In particular...
>But I don't believe they are actually judged on their ability to taste wines and detect flavours. Designing such tests would be simple - rate of successful flavour identification based on data from other Somemeliers.
That is exactly how the tasting portion of the exam works. And
>In my opinion they judge it as a dramatic performance: how quickly can the candidate rattle-off various flavours? How "high class" is the language they use to describe the flavours?
This isn't really how it works, especially the "high class" language comment. Actually there are a variety of different things that they are supposed to identify about the wine, known as "the grid" [1]. You can take a look at it yourself. They are essentially judged based on how well their grid matches up with the consensus grid from the master sommeliers.
This is something I notice in advertising generally: whatever they most emphasize is least likely to be true about the product (e.g., the "great taste" of McDonald's).
The other day I looked up the wordle answer (I know I know). The first result was a site where I had to scroll through about 19 paragraphs of SEO vomit to get to the answer. The page could literally have one word on it and serve it's purpose. If that isn't a sign that the the internet, or at least google search, is dead, I don't know what is.
There's a whole industry of sites like this for NYT crossword answers, to the point that it's often impossible to get any organic results at all for something that's been clued in the Times, which is frustrating because I don't just want the answer (there's a button in the app that does that already), I want to learn about the thing I’m unfamiliar with. Luckily, most topics that come up have some content on Wikipedia so I go there instead.
Ugh yes that’s super annoying! I don’t want the answer spoon-fed to me. The point of searching it is to actually learn a little something which will help me remember in the future
lol the same happens with WSJ. It’s very disappointing to try and do a bit of research on a clue and get the answer on a bot site linked to the puzzle you’re doing. Interesting how they do it though.
It’s like a DDoS attack on your mind (distributed because everyone is doing it). The attention span economy at its finest.
Reminds me of those ways to catch spammers or bots by occupying some of their resources with meaningless tasks for as long as possible. Except it’s turned around.
Sometimes I wonder if there’s even any real money in ads anymore or if it’s just a giant circle jerk that slowly destroys society…
Speaking of bots, I'd be interested to know the percentage of articles on major traffic content sites are authored or co-authored by AI.
My suspicion is this is rife given how many articles read poorly and are almost entirely fluff. If this is true it would appear we are doomed to algorithms shaping our online experiences, which is worrying given the existing shrinking diversity of opinion and content. It's like a entropic gene pool in nature, but with information.
Sports stories are frequently written by bots/AI/whatever you want to call them. Sports stats makes it relatively easy to create written text articles about games. When first rolled out, there were some obvious tells such as an excessive amount of "for the first time this season" or "set a record for the season" in articles about first games of the season. Unusual plays or quirky behavior by participants tend to be missed in these articles but otherwise they're serviceable though somewhat dry and bland.
Stock analysis articles are very commonly generated automatically from templates. The selected template is based on some real characteristics of a described company (whether the stock went up or down recently, what is the P/E etc.), but the content is generic and reused across all companies with similar characteristics.
Apparently, using a machine translation as a basis and working through correcting it to read or write foreign languages is a growing trend and that’s a form of computer assisted literacy if I were to guess.
Bar that, it’s humans outrunning AI in the race to the bottom with a head start. Human people can be forced to be incredibly machine like.
> Human people can be forced to be incredibly machine like.
Amazon has a site for humans to pick up small tasks and get paid for them called mechanical turk [1], which is a reference to a fake chess playing machine with a human inside [2]. With the great resignation and the workforce otherwise pushing for a reasonable standard of living, I'm not sure how heavily mturk is still used as depending on tasks and speed it's really sub-minimum wage work for many people. [3] But as The Atlantic article says, sometimes it's the only work people can get.
I think the quality of searches like “best TV” will improve dramatically once language models are used to generate SEO spam. Anything would be an improvement over what today’s human spam bots produce.
Lately, half the results I get are pages that I cant view without paying to some service or signing up for a free trial.. Google literally serves up results that are unreachable ..
Can we also talk about how Google allows top organic results to LinkedIn, Instagram, and Facebook which literally present you with a LOGIN page before you see anything related to your search?
Absolute worst UX ever, yet they allow it because it's their SV buddies.
> yet they allow it because it's their SV buddies.
At one time, Google used to ban sites for "cloaking": offering one version of the page to the crawler, and another to the user.
But over time they got into trouble from these sites, getting sued left and right. Being accused of putting up a wall, and abusing their monopoly. Eventually, these sites won out and Google dropped this requirement.
Similar story with Image search. "Blah blah something something copyright something". And voila - now you can't get direct access to images on the search results page.
This might be the case for news sites which show limited content until you sign into a paywall, but I don't recall the biggest offenders which show NO relevant content at all (LinkedIn and Facebook) saying anything.
Classic example of this kind of content... Try searching "How to use X to get stains out of Y".
You will find a page for almost any X and Y combination. And they will all have wording like "Put some X on the stained Y... wait a bit... rub it in... and then put it through the washing machine. Hope it works!".
I mean then is DIT really a conspiracy theory? I know that people in HN have already started doing replies with AI in some threads. (only because they tell people). Wait a few more years and there will be no way to know if forums are just bots generating content.
I guess HN would be a weird outlier, because there are no ads. Except maybe the bots would be useful after all when various "Show HN" or "Launch HN" and have the bots cheer those companies and get random publicity.
The idea that a genuinely Dead Internet might be an improvement over the current internet experience is a fun one, and one I don't entirely disagree with.
Within the last 6 months I’m seeing a rise of very good NLP content farms that almost have what I’m looking for but not quite. I think what they do is start off a transformer based NLP with example queries that others have searched for and it generates a realistic looking answer that’s usually wrong. The scale and breadth of these could only be done by machines.
Writing SEO content isn't necessarily one of them anymore. Check out AI content services like https://www.frase.io/. You can also generate things like product descriptions inside the GPT-3 playground.
Reminds me of an old The Parking Lot Is Full comic.
> Little-known Fact #839: There are only twenty-three people alive today, and you're one of them; everyone else you know just looks human to lull vou into not searching for the other twenty-two. Lonely? You should be.
I would expect an automaton to be able to spell; so perhaps the presence of spelling errors is a mark of an authentic page. Maybe one could force Goo to spit out authentic results by including a strategically-misspelled word in the search terms.
The most important factor was cramming SEO terms and links to keep people on the website into the articles.
The result is trashy articles that could well have been written by a bot but aren't. This could possibly be done with the help of curated bot-content, but I think we're far away from the point where this is really more profitable than getting students to do the work.
It's people but they work like bots.