Thanks to ChatGPT, the pure internet is gone. Did anyone save a copy?
In the post-nuclear age, scientists noticed a peculiar problem: steel produced after 1945 was contaminated. Atomic bombs had infused the atmosphere with radioactivity, which contaminated the metal.
This made most steel useless for precise equipment such as Geiger counters and other highly accurate sensors. The solution? Salvage old steel from sunken pre-war battleships resting deep on the ocean floor, far away from the nuclear fallout. This material, known as low-background steel, became prized for its purity and rarity.
Fast forward to 2025, and a similar story is unfolding — not under the sea, but across the internet.
Since the launch of ChatGPT in late 2022, AI-generated content has exploded across blogs, search engines, and social media. The digital realm is increasingly infused with content not written by humans, but synthesized by models and chatbots. And just like radiation, this content is tricky for regular folks to detect, is pervasive, and it alters the environment in which it exists.
This phenomenon poses a particularly thorny problem for AI researchers and developers. Most AI models are trained on vast datasets collected from the web. Historically, that meant learning from human data: messy, insightful, biased, poetic, and occasionally brilliant. But if today's AI is trained on yesterday's AI-generated text, which was itself trained on last week's AI content, then models risk folding in on themselves, diluting originality and nuance in what's been dubbed " model collapse."
Put another way: AI models are supposed to be trained to understand how humans think. If they're trained mostly on their own outputs, they may end up just mimicking themselves. Like photocopying a photocopy, each generation becomes a little blurrier until nuance, outliers, and genuine novelty disappear.
This makes human-generated content, from before 2022, more valuable because it grounds AI models, and society in general, in a shared reality, according to Will Allen, a vice president at Cloudflare, which operates one of the largest networks on the internet.
This becomes especially important as AI models spread into technical fields, such as medicine, law, and tax. He wants his doctor to rely on content based on research written by human experts from real human trials, not AI-generated sources, for instance.
"The data that has that connection to reality has always been critically important and will be even more crucial in the future," Allen said. "If you don't have that foundational truth, it just becomes so much more complicated."
Paul Graham's problem
This isn't just theoretical. Problems are already cropping up in the real world.
Almost a year after ChatGPT launched, venture capitalist Paul Graham described searching online for how hot to set a pizza oven. He found himself looking at the dates of the content to find older information that wasn't " AI-generated SEO-bait," he said in a post on X.
Malte Ubl, CTO of AI startup Vercel and a former Google Search engineer, replied, saying Graham was filtering the internet for content that was "pre-AI-contamination."
"The analogy I've been using is low background steel, which was made before the first nuclear tests," Ubl said.
Matt Rickard, another former Google engineer, concurred. In a blog post from June 2023, he wrote that modern datasets are getting contaminated.
"AI models are trained on the internet. More and more of that content is being generated by AI models," Rickard explained. "Output from AI models is relatively undetectable. Finding training data unmodified by AI will be tougher and tougher."
The digital version of low-background steel
The answer, some argue, lies in preserving digital versions of low-background steel: human-generated data from before the AI boom. Think of it as the internet's digital bedrock, created not by machines but by people with intent and context.
One such preservationist is John Graham-Cumming, a Cloudflare board member and the company's CTO.
His project, LowBackgroundSteel.ai, catalogs datasets, websites, and media that existed before 2022, the year ChatGPT sparked the generative AI content explosion. For instance, there's GitHub's Arctic Code Vault, an archive of open-source software buried in a decommissioned coal mine in Norway. It was captured in February 2020, about a year before the AI-assisted coding boom got going.
Graham-Cumming's initiative is an effort to archive content that reflects the web in its raw, human-authored form, uncontaminated by LLM-generated filler and SEO-optimized sludge.
Another source he lists is "wordfreq," a project to track the frequency of words used online. Linguist Robyn Speer maintained this, but stopped in 2021.
"Generative AI has polluted the data," she wrote in a 2024 update on coding platform GitHub.
This skews internet data to make it a less reliable guide to how humans write and think. Speer cited one example that showed how ChatGPT is obsessed with the word "delve" in a way that people never have been. This has caused the word to appear way more often online in recent years. (A more recent example is ChatGPT's love of the em dash — don't ask me why!)
Our shared reality
As Cloudflare's Allen explained, AI models trained partly on synthetic content can accelerate productivity and remove tedium from creative work and other tasks. He's a fan and regular user of ChatGPT, Google's Gemini, and other chatbots such as Claude.
And just like human-generated data, the analogy to low-background steel is not perfect. Scientists have developed different ways to produce steel that use pure oxygen.
Still, Allen says, "you always want to be grounded in some level of truth."
The stakes go beyond model performance. They reach into the fabric of our shared reality. Just as scientists trusted low-background steel for precise measurements, we may come to rely on carefully preserved pre-AI content to gauge the true state of the human mind — to understand how we think, reason, and communicate before the age of machines that mimic us.
The pure internet is gone. Thankfully, some people are saving copies. And like the divers salvaging steel from the ocean floor, they remind us: Preserving the past may be the only way to build a trustworthy future.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
WIRED
2 hours ago
- WIRED
Samsung Teases Z Fold Ultra, Bing Gets AI Video, and Nothing Sets A Date—Your Gear News of the Week
Plus: Ruark has new speakers, Photoshop comes to Android and summer's finest music player gets updated. All products featured on WIRED are independently selected by our editors. However, we may receive compensation from retailers and/or from purchases of products through these links. Bing has added a new AI-powered video generation tool to its mobile app, that's built on OpenAI's Sora text-to-video model. That's a feature that, even now, is exclusive to ChatGPT subscribers—but Bing users will get it for free. The vertical video creations are 5 seconds long but aren't generated instantly—once you type in a prompt, you'll get a notification when the video is ready. The Standard generation speed is free, but you'll also be able to access the 'Fast' option 10 times before you'll need to cough up 100 Microsoft Reward points to keep using it at that speed. You can share these videos anywhere, and they'll be stored in the Bing app for 90 days. The video generation wars have been heating up over the last year. Google debuted its Veo 3 model at Google I/O in May, with significant upgrades to quality. Chinese phone brand Honor also recently partnered with Google to add a feature that converted still images in the Gallery app into 5-second video clips through Google's Veo 2 model. With the ability to now generate videos at our fingertips, it'll make it easier than ever to share exactly what you're envisioning to a friend or loved one, but it'll be even harder to distinguish what's real and what's not. Nothing Sets a Date for Phone (3) and Headphone (1) London-based Nothing took a year-long break from its top-end smartphone line after it debuted the Phone (2) in 2023. In that time, it created the Phone (2a) in 2024, which went on to be one of the company's best-selling handsets. There's already a successor for those budget phones—the Phone (3a) series—but now it's time for a new flagship from the brand. The company announced this week that it will unveil the Phone (3) at an event in London on July 1 at 1 pm ET. This content can also be viewed on the site it originates from. We have a few details so far. The phone may not have the Glyph light interface on the back anymore, though it seems like Nothing has cooked up a new dot matrix light pattern instead. The company says it'll be its first true flagship phone with premium materials, and it'll have a high price to boot: somewhere around £800. But the spotlight won't just be on a new phone. This week, Nothing also shared that it will be entering 'a new product category" at the event with its first-ever pair of headphones. Creatively dubbed Headphone (1), it'll be Nothing's first over-ears, but follows a long line of wireless earbuds. Not too long ago, Nothing announced a partnership with iconic audio brand KEF. Perhaps these headphones will be the pair's first collab. Samsung Teases a Galaxy Z Fold Ultra Samsung's Galaxy Unpacked event is also expected to take place in July, and rumors abound that we'll see the Galaxy Z Fold7, Galaxy Z Flip7, and even a Flip 7 FE—a cheaper version of the company's flip folding phone. But Samsung took time to tease something else: an Ultra variant of its folding phone. Or so we think. In a blog post on Samsung Newsroom, the company vaguely talks about a folding device that can match the capabilities of its existing Ultra phones, like the Galaxy S25 Ultra. What remains unclear is if the upcoming Galaxy Z Fold7 will offer an Ultra-like experience with no compromises, or if there will indeed be a dedicated Ultra version of that phone. Until now, there have been trade-offs between the Fold phones and Samsung's Galaxy Ultra phones, with the latter offering a nicer camera experience, better battery life, and other perks like the stylus. Perhaps Samsung has found a way to replicate the true Ultra experience on its next generation of the Fold. The company has a teaser video showing the silhouette of the Fold opening and closing. There have been rumors that Samsung is working on a tri-fold phone, like Huawei's version that nets you an even bigger screen when unfolded; you'd think if anything got the Ultra moniker, it'd be that device. We'll have to wait and see. Ruark's MR1 Mk3 Get Some Serious Upgrades The Ruark MR1 have been some of the best sounding, most stylish desktop stereo speakers you can buy at their price for over a decade. Now in their third generation, they have been rebuilt from the ground up, with the aim of improving sound quality, refining the hand-crafted design and adding in some great new features to make them even more versatile than before. This includes adding aptX HD playback for higher quality Bluetooth sound, a USB audio connection for easy high-resolution playback and a moving magnet phono stage for powering a turntable. The petite package is available now, and costs $579/£399. — Verity Burns — Verity Burns Photoshop for Android Is Here Adobe has finally released Photoshop for Android. No, this isn't Photoshop Express or Photoshop Touch—previous, largely failed attempts at bringing Photoshop to mobile. Photoshop for Android mirrors the version of Photoshop for iPhone released earlier this year. You can download the public beta for Android today. The mobile app has nearly everything you'll find in Adobe's desktop version, including layer-based editing and tools like masks, clone stamp, intelligent selection options, and all the tone and curve adjustment tools. The user interface is radically different, but Photoshop veterans will likely get the hang of the mobile version quickly. I've been testing the Android app for a couple of days now, and it's fairly impressive, but a few things are missing. The biggest for me is the ability to crop by pixels rather than ratio, which seems like a very odd limitation. Content-aware fill is also still "coming soon." Adobe has been heavily touting the AI features, which make it possible to do smart selections that would be difficult otherwise. I've found this feature works like on desktop (it relies on the same cloud backend), but I still don't have much use for it. — Scott Gilbertson Poolsuite V3 Has Your Summer Playlist Sorted "Throw your laptop out the damn window and drag that 1994 Kawasaki 750SX stand-up jet ski out of Uncle Pete's garage, because summer is officially here." This is how Poolsuite, possibly the finest curated music app for outdoor frivolity, announces the arrival this week not only of a throughly revamped and upgraded version of its already superb iOS media player, but also that it's finally available on Android as well. This perfectly judged throwback tone pervades throughout the app, which now adds hundreds of new tracks across seven channels, as well as mobile mixtapes to go with the aesthetic overhaul. Sun-drenched playlists lovingly curated to lift spirits and deliver virtual vitamin D for free. If you haven't downloaded it already, do so right now—and never worry about what tunes to play at a BBQ ever again. — Jeremy White The New Hublot Big Bang Unico Summer 2025 Continuing the summer theme in style is this new limited edition beach-ready Big Bang from Hublot. 'As light as a sea breeze with its featherlight ceramic,' says the brand, with a micro blasted 'orange case that glows like the golden hour.' Well, I tried it on at Watches & Wonders in April, and unlike some other darker hued versions of this watch, it's playful and thoroughly approachable, yet with 100 meters of water resistance is equally at home either at a pool party or in deep waters. A one-click system also allows the included three interchangeable white rubber-lined straps in sky blue, dark blue or orange to be swapped in a jiffy, and the 72-hour power reserve keeps things going when off the wrist. The price? $31,300 (£26,900) but only 100 will be released. — Jeremy White
Fast Company
3 hours ago
- Fast Company
This free AI supersite is like Gemini Deep Research on steroids
Everywhere you look these days, there it is—some manner of breathlessly hyped new 'AI' service that's, like, totally gonna change your life forever. (Like, totally. For realsies.) Or so they say. In reality, of course, most of this stuff is far more fallible, limited in utility, and inadvisable to use outside of super-specific scenarios than most tech companies (and self-declared 'AI gurus') would lead you to believe. But AI, in its current form, isn't entirely useless. Far from it, in fact: This type of tech can be quite helpful in the right sort of scenario and, critically, if you think about it in the right way—not as an end-all instant answer machine but as a starting point for certain types of specific tasks or info-seeking. And as we wade our way through a year that's absolutely overflowing with overwrought AI ballyhoo, I've got just the tool for you to sift through that sea and seek out some surprising shiny pearls amid all the overwhelming noise. Be the first to find all sorts of little-known tech treasures with my free Cool Tools newsletter from The Intelligence. One useful new discovery in your inbox every Wednesday! Deep research, done right So, you've probably heard all about ChatGPT, Gemini, Perplexity, and the likes, right? They're all generative AI chatbots, which means they use a snazzy-sounding word prediction engine to analyze language patterns and answer your questions, among other more ambitious tasks. 🔎 One of their biggest recent advancements is the ability to perform what everyone's calling 'deep research'—a fancy way of saying they'll dive deep into a topic for you and create a detailed report of info, almost like a custom-made dossier, based on knowledge from all over the web. Again, I can't emphasize enough: The info here isn't infallible. These systems can—and do—get stuff wrong and sometimes even flat-out make up nonsense out of thin air. 🧠 But, as a starting point—especially when they include links to their sources so you can confirm info on your own and use it as an entryway to research as opposed to the final product—it really can save you time and give you a great way to get into a complex topic. And the tool I want to show to you today makes that feature far more powerful, useful, and also affordable than it's ever been before. ⌚ It'll take you 20 seconds to try out for yourself. ➜ It's called, amusingly, Ithy. (Try saying that 10 times fast!) And all it does, in a nutshell, is bring together the 'deep research' tools from a slew of different AI engines—including ChatGPT and Google's Gemini along with Perplexity, Meta AI, and more—into a single streamlined prompt. That means you can use 'em all together to create a single super-report on any subject imaginable. ✅ It couldn't be much easier to make happen, either: First, open up Ithy in any browser, on any device you're using. Type your question or the subject you're thinking about into its box and tap or click the arrow icon within that same line to get going. Select either 'Fast,' if you don't feel like waiting, or 'Deep,' if you've got time and want this thing to go especially in-depth. (Even the 'Fast' path is pretty darn deep, if you ask me.) And, well, that's about it. Ithy will think for a bit, then serve up an impressively detailed dossier on whatever it is you requested—with info coming from a mix of all those AI engines, combined and seamlessly blended together. And I mean seriously detailed, too—with all sorts of sections, graphics, FAQs, and external links for original sources so you can do your own reading and see exactly where it got its info. 📌 Here's a link to the sample report shown here, if you want to look even more closely. ☝️ Now, for the especially cool part: Ithy lets you do all of this free of charge —up to a point. The site gives you five report-creating credits to start, even if you don't sign in. Once you create an account (for free), you'll get 10 credits per month and can optionally then bump up to an unlimited Pro plan—which includes access to the typically pricey pro levels of Gemini and OpenAI—for seven bucks a month, if you go for the annual setup. But even if you don't go that route, 10 in-depth reports per month from all the web's leading AI engines together is a pretty powerful perk to have at your fingertips, without so much as dropping a dime. Ithy is entirely web-based —no downloads or installations required. It's free for up to 5 reports total or 10 reports per month, if you create an account—and optionally available in $7-per-month (paid annually) or $20-per-month (paid monthly) plan for its fully featured, limit-free Pro version. Like most AI engines, Ithy does use questions submitted to its site as training to further improve its AI systems. The questions are also being shared with the associated third-party AI sites, of course. So you'll want to think carefully about what you ask and avoid sending anything especially sensitive or personal (but really, it's designed to answer questions and provide info, so hopefully you wouldn't be submitting your banking info and Social Security number, anyway!).
Business Insider
3 hours ago
- Business Insider
AI search's user experience may be the best it'll ever get, says one founder
By day, Lily Clifford is the CEO and founder of Rime Labs. The startup creates the voice on the other end of the line when you call to order from restaurants like Domino's or Wingstop. Rime trains AI models to create voices with specific regional accents, tones, and other elements that make them easier to converse with. Clifford also uses AI in her daily life, especially in lieu of search engines, she told Business Insider. Instead of pulling up a search engine when she has a question, Clifford usually turns to generative AI chatbots like OpenAI's ChatGPT or Google's Gemini. She said the experience reminds her of using Google or other search engines in the late 1990s and early 2000s. That's when she thinks the user experience was at its prime. "My hot take here is these applications might be the best that they ever will be," she said. Search engines used to be simpler, Clifford said. There were far fewer ads and sponsored results. And optimizing webpages to get more clicks — a practice known as SEO — was in its infancy. Those developments spawned new businesses and became features of the modern internet. But Clifford said search results have also gotten worse for users. It's common to see multiple sponsored results above more relevant ones in a search, for instance. AI chatbots, meanwhile, haven't gone through the same evolution — yet. Companies and individuals are still experimenting with usinggenerative AI for lots of tasks, from writing emails to creating images for advertising campaigns. Many people, like Clifford, use AI as a replacement for search engines. Ask AI a question, and it will often give you an answer in just a few sentences. For some, that's more appealing than clicking through several results from a search engine until you find the information that you're looking for. AI search results can also give users contradictory or incorrect information, though, creating a potential downside to the quick-and-easy answers. Still, Clifford noticed the user experience gap between the chatbots and search engines during a recent trip to Milan, she said. While there, she used an AI chatbot to look for a local place to buy a silk blouse. The chatbot pointed her toward a local seamstress who sold custom blouses through Instagram. "It wasn't like 'Go to Forever 21,' which is probably what would've happened if I typed it into Google," she said. "It was totally wild and fun to use." But, Clifford thinks it's a matter of time before AI chatbots go the way of the search engines before them. Some companies with big investments in generative AI search tools are taking steps in that direction. Last month, Google said it would expand its use of ads in some of the AI Overviews that appear at the top of its search results, for example. And some marketing experts now offer help with " answer engine optimization," or AEO.
