The Dead Internet Theory May Be Coming True, Pew Research Findings Show
Are you reading more text written by an AI bot? A Pew Research study published this week says it’s very likely.
That’s not breaking news if you’re aware of the Dead Internet Theory that dates back to 2021. CNET writer Trisha Jandoc dove into that theory last year, which basically says people believe the internet died 10 years ago in 2016 and that much of what we see online now is AI-generated.
There is some truth to that. Pew researchers scanned nearly half a million English webpages from the past five years using Common Crawl, a web archive. That dates back to before ChatGPT was made publicly available in November 2022. Researchers ran those pages through Open Pangram, an AI detection tool, to see if it was written or edited by AI. Then the researchers looked at a sample of 10,000 pages from last month to spot signs of AI-written content. They looked for phrases, words and language that are more common among AI bots than humans.
Researchers made it clear that AI detection models don’t always get it right. Sometimes they flag text written by humans as AI authorship and AI-written text as being written by a human.
Nonetheless, the study found that 10% of the samples “show significant signs of AI authorship.”
(Score: 5, Interesting) by gnuman on Tuesday August 25, @08:04AM (15 children)
The more you fill your time, with AI or with anti-social networks, or just regular socializing, the less stuff are you going to write on your blog. Heck, these days people probably write less in general and upload video of themselves talking instead.
It's not AI as much as people's lack of time to write or even read.
(Score: 5, Insightful) by bzipitidoo on Tuesday August 25, @10:32AM (8 children)
My lack of time drives me away from video and towards text. I can read much faster than a talking head can talk.
(Score: 3, Funny) by Anonymous Coward on Tuesday August 25, @02:21PM (7 children)
Increase playback speed to 4x.
(Score: 4, Touché) by Freeman on Tuesday August 25, @02:46PM (6 children)
I can skim an article and ascertain whether it's worth my time much faster than I could do the same with a video. Even at 4x speed on video. Sure, you can skip through with 4x speed going I guess, but you will not get the same experience skimming a video as skimming a wall of text.
Joshua 1:9 "Be strong and of a good courage; be not afraid, neither be thou dismayed: for the Lord thy God is with thee"
(Score: 5, Insightful) by Unixnut on Tuesday August 25, @04:08PM (5 children)
I can skim read faster than I can watch a 4x video, plus I find it hard much beyond 2x to understand what the person is saying, already at 2x voices start to become a bit garbled in places.
Reading has the advantage is that I can jump forward and back skipping words, sentences or even paragraphs to get the gist of what the text contains. I can't do that with video, unless I want to pause, click back a few times, try to find the rough spot where the subject said what I wanted to hear, realise they said it earlier, go further back, etc....
No matter how you cut it, information density and speed of ingestion is faster with text than with video, which is why I prefer reading to watching videos in the vast majority of cases (Exclusions are for demonstration, when I need to see someone do something for the sake of understanding the process).
(Score: 2, Interesting) by Anonymous Coward on Tuesday August 25, @04:25PM
> Reading has the advantage is that I can jump forward and back skipping words, sentences or even paragraphs to get the gist of what the text contains.
Good description of what humans do (or at least what I try to do...)
When I've asked Gemini to summarize a long article or email discussion, it does a poor job--regularly skipping important parts of the text and/or connecting parts of the text that should be kept separate.
(Score: 3, Interesting) by corey on Tuesday August 25, @09:00PM (3 children)
> No matter how you cut it, information density and speed of ingestion is faster with text than with video
I get your point, which is valid, but I don’t really agree.
A video is a set of moving pictures. A picture tells a thousand words. Cliche aside, I find it faster to watch someone demonstrate with the use of diagrams and videos, complex physical or engineering things. Like a lecturer with a whiteboard. Reading the same in a book is slow for me and more difficult. And learning another language is best in person or with videos. Product reviews, same, I like to see the product and see it moved and interacted with.
But yeah for story telling or news or whatever, text is best for me. I only use videos for particular purposes as above.
(Score: 5, Insightful) by bart9h on Wednesday August 26, @01:20AM
Your "picture tells a thousand words" does not work if it is a video of someone talking. For that, reading is indeed faster.
(Score: 3, Insightful) by Freeman on Wednesday August 26, @01:31PM (1 child)
In the case of a video versus a script of said video. The script is the picture of the video.
Joshua 1:9 "Be strong and of a good courage; be not afraid, neither be thou dismayed: for the Lord thy God is with thee"
(Score: 3, Insightful) by Freeman on Wednesday August 26, @01:33PM
Beyond that, a video is typically made with a different mindset than an article. A good article is generally much better at conveying all the details. The video is for those that can't read.
Joshua 1:9 "Be strong and of a good courage; be not afraid, neither be thou dismayed: for the Lord thy God is with thee"
(Score: 4, Interesting) by PiMuNu on Tuesday August 25, @11:04AM (1 child)
> lack of time to write or even read.
I spend far more time reading and writing than anything else.
The stuff I write is usually extremely high cost i.e. academic papers. Typically a year's work to write a ten page report.
(Score: 3, Informative) by bzipitidoo on Tuesday August 25, @05:58PM
I know that well. Last time I did a research paper, spending months on it, it was rejected. I tried one more time in a different publication. Rejected again. I abandoned it. And I abandoned a notion I had of applying for a position in academia. I don't have a position in academia, and I asked myself, what the F am I doing? In the world of Publish or Perish, I am long perished. I don't have any position to keep, so I certainly don't need to do research for that purpose. Partly, I was doing it for my own amusement. But I thought it good enough to be worth publishing, so I gave it a shot. All that work to knock the raw findings and data into a publishable form, just to have some reviewers misunderstand it and tell me it's not good enough. They wrote as if it is customary that researchers assemble a large team of authors, who have hordes of grad students they can put to work on relevant experiments. I have no such resources. They also bragged about having rejected more than half the submissions they received, seemingly to show what high standards they have. The reviewers didn't say it was wrong. But one of them said it was "bikeshedding". I put it out on Arxiv anyway. So far as I know, no one has ever read it.
I had more reasons than that rejection for giving up on academia. We've been living through a time of much hostility to education, which has manifested in poor pay, and greater, uglier, and more cutthroat competitiveness, less respect, and worsening conditions. For all I know, journals have cooperated with a desire to limit the number of researchers by being excessively demanding. I find it ridiculous that people could go through the years of education and inspiration it takes to earn a PhD, and then have to face such high rejection rates, as if they haven't had to prove themselves many times over to earn the degree!
(Score: 4, Interesting) by JoeMerchant on Tuesday August 25, @01:41PM (2 children)
I whipped off a new requested tool implementation yesterday with an LLM agent in about 2 hours. Straightforward simple thing, 2K install package, 150 lines of README about the tool and how to use it, 150 lines in the functional tool itself, and 200 lines in the script that makes the install package.
Without LLM assistance, the README would have been 50 lines, lacking clarifying detail that may or may not be helpful in the future, the tool may have been 30 lines but it would have covered fewer edge cases, had less effective error handling, mostly been a "here's the happy path, tell me if anything goes wrong" thing. And I probably wouldn't have scripted the packager at all - more's the shame, because that packager is going to make future revisions a 5 minute snap instead of a 2 hour research project. I probably would have spent 6-8 hours bringing this thing up to a functional state without LLM assistance and just cut the effort off there rather than continuing to bring the additional polish to the project. Some of the polish is overkill, but IMO it _is_ adding value and will potentially save future effort, it is just that the calculus has changed - old math: do I spend an additional 4 hours polishing this turd for a 5 percent chance of saving 6 hours in the future once or twice? New math: accept the default output of the LLM, save the 20 minutes it would have taken to prune stuff that actually might have a chance of being valuable in the future, or go ahead and accept everything that looks like it might have a net positive value greater than the cost of reading through it?
Oh, about half of that 2 hour LLM assisted project was spent in refining the output, removing useless / distracting junk, refining the design to keep debug information away from the users (put it in the error log files) that was split about 50/50 between me having given less than optimal instructions in the first place and reigning in the LLM's enthusiasm for things we don't need. Still saved me at least 4 hours of work and produced a better end product.
🌻🌻🌻🌻✌️ [google.com]
(Score: 2) by VLM on Wednesday August 26, @02:27PM (1 child)
I would point out that another aspect of the new math is you were personally responsible and reputationally liable for 50 lines of README and now you're personally responsible and liable for 150 lines.
Back in the old days, pre 1980s, secretaries would spend an hour hand typing an mimeographing a memo to ten people. In the 1980s desktop publishing era, that memo could be created and printed in 15 minutes. Unfortunately, what we got in the DTP era was now two hours of labor, selecting 15 different fonts and clipart to send the memo to the original ten people plus 90 more who have their time wasted because they don't need it.
(Score: 2) by JoeMerchant on Wednesday August 26, @03:54PM
> you were personally responsible and reputationally liable for 50 lines of README and now you're personally responsible and liable for 150 lines.
It's true, though those 150 lines actually cover my ass better than the 50 did, explain the edge cases better (or at all). Reviewing the 100 extra lines takes little time to ensure they're reasonably correct, authoring them in the first place would have tripled my lookup and verification time to get the content in the first place - once it's looked up it's relatively easy to verify - either at a glance, or sometimes these things warrant testing.
> send the memo to the original ten people plus 90 more who have their time wasted because they don't need it.
You should see our Corporate training requirements... I literally spend more time training than I used to spend commuting (though with the commute I also had to do the training.) And 90%+ of the training is worthless / inane / inapplicable / something I might do someday but haven't yet in the past 15 years...
🌻🌻🌻🌻✌️ [google.com]
(Score: 2) by mcgrew on Wednesday August 26, @07:27PM
Well, the internet won't ALL be AI garbage as long as my non-AI garbage sites are on the internet.
I seem to have a knack for spotting AI, although some takes longer for me to notice. A huge amount of YouTube is AI garbage. Watch gestures, AI is awful at getting human gestures right. One was REALLY obvious, a judge at his bench talking about a case, and his hand went right through his face more than once. Another not quite so obvious one is an Asian "man" that talks about tech; it's not a man, it's an AI cartoon.
When you see moving movie scratches on a still photo, it's either AI or an ignorant young human producing it who thinks that makes it look cool..
Donald Trump is a mixture of Warren Harding, Calvin Coolidge, and Herbert Hoover, multiplied by ten.
(Score: 4, Insightful) by VLM on Wednesday August 26, @02:22PM
One propaganda technique to discredit an opinion is to make a false extremist statement about it which everyone would disagree with.
Its more like individual contribution has been steadily decreasing for decades replaced with fake farms of corporate freelancer produced slop, which was more expensive than AI generated freelancer produced slop...
There WAS an entire industry around paying freelancer humans to create fake cooking recipe websites but that's exploded and now there's 10x as many completely AI generated slop sites all choked with Google Adsense ads.