Established publishers seeking relief from the whims of social-media platforms and a brutal advertising environment found in product recommendations steady growth and receptive audiences, especially as e-commerce became a more dominant mode of shopping. Today, these businesses are materially significant — in a 2023 survey, 41 percent of surveyed media companies said that e-commerce accounted for more than a fifth of their revenue, which few can afford to lose. It is a relatively new way in which publishers have become reacquainted — after social-media traffic disappeared and “pivots to video” completed their rotations — with queasy feelings of dependence on massive tech companies, from Facebook and Google to Amazon and, well, Google.

Time magazine announced a brand called Time Stamped, “a project to make perplexing choices less perplexing by supplying our readers with trusted reviews and common sense information,” with “a rigorous process for testing products, analyzing companies,” and making recommendations. In early 2024, the Associated Press announced its own recommendation site, AP Buyline, as an “initiative designed to simplify complex consumer-made decisions by providing its audience with reliable evaluations and straightforward insights,” based on “a thorough method of testing items, evaluating companies and suggesting choices.” Both sites currently recommend money-related products and services, including credit cards, debt-consolidation loans, and insurance policies, categories that can command very high commissions; the AP reportedly plans to expand to home products, beauty, and fashion this month.

Time Stamped and AP Buyline share strikingly similar designs, layouts, and sensibilities. Their content is broadly informative but timid about making strong judgments or comparisons — an AP Buyline article about “The Best Capital One Credit Cards for 2024” heartily recommends nine of them. The writer credited for the article can also be found on Time Stamped writing about Chase credit cards, banks, and rental-car insurance. On both sites, if you look for it, you’ll also find a similar disclaimer. For Time: The information presented here is created independently from the TIME editorial staff. For the AP: AP Buyline’s content is created independently of The Associated Press newsroom. By independently, both companies mean that their product-recommendation sites are operated by a company called Taboola.

Over the years, Taboola, which is best understood as an advertising company, became a major player in affiliate marketing, too, through its acquisition of Skimlinks, a popular service for adding affiliate tags to content. In 2023, it started pitching a product called Taboola Turnkey Commerce, which claims to offer the benefits of starting a product-recommendation sub-brand minus the hassle of actually building an operation.

As her site has disappeared from view on Google, Navarro has been keeping an eye on popular search terms to see what’s showing up in its place. Legacy publishers seem to be part of Google’s plan, but a recent emphasis on what the company calls “perspectives” could also be in play. Reddit content is getting high placement as it contains a lot of conversations about products from actual customers and users. As its visibility in Google has increased, though, so has the prevalence of search-adjacent Reddit spam. Since the update has started rolling out, Navarro says, she has “seen a lot of generic review sites” getting ranked with credible-sounding names, .org domains, and content ripped straight from Amazon reviews.

“You can search all day and learn nothing,” she says. “It’s like trying to find information inside of Walmart.”

For now, Navarro is unimpressed with these AI experiments. “It’s just shut-up-and-buy,” she says — if you’re doing this search in the first place, you’re probably looking for a bit more information. In its emphasis on aggregation, its reliance on outside sources of authority, and its preference for positive comparison and recommendation over criticism, it also feels familiar: “Google is the affiliate site now.”

Google is trying to kill the 10 blue links. Twitter is being abandoned to bots and blue ticks. There’s the junkification of Amazon and the enshittification of TikTok. Layoffs are gutting online media. A job posting looking for an “AI editor” expects “output of 200 to 250 articles per week.” ChatGPT is being used to generate whole spam sites. Etsy is flooded with “AI-generated junk.” Chatbots cite one another in a misinformation ouroboros. LinkedIn is using AI to stimulate tired users. Snapchat and Instagram hope bots will talk to you when your friends don’t. Redditors are staging blackouts. Stack Overflow mods are on strike. The Internet Archive is fighting off data scrapers, and “AI is tearing Wikipedia apart.”

it’s people who ultimately create the underlying data — whether that’s journalists picking up the phone and checking facts or Reddit users who have had exactly that battery issue with the new DeWalt cordless ratchet and are happy to tell you how they fixed it. By contrast, the information produced by AI language models and chatbots is often incorrect. The tricky thing is that when it’s wrong, it’s wrong in ways that are difficult to spot.

The resulting write-up is basic and predictable. (You can read it here.) It lists five companies, including Columbia, Salomon, and Merrell, along with bullet points that supposedly outline the pros and cons of their products. “Columbia is a well-known and reputable brand for outdoor gear and footwear,” we’re told. “Their waterproof shoes come in various styles” and “their prices are competitive in the market.” You might look at this and think it’s so trite as to be basically useless (and you’d be right), but the information is also subtly wrong.

It’s fluent but not grounded in real-world experience, and so it takes time and expertise to unpick.

Over the past six months, I spoke with more than two dozen annotators from around the world, and while many of them were training cutting-edge chatbots, just as many were doing the mundane manual labor required to keep AI running. There are people classifying the emotional content of TikTok videos, new variants of email spam, and the precise sexual provocativeness of online ads. Others are looking at credit-card transactions and figuring out what sort of purchase they relate to or checking e-commerce recommendations and deciding whether that shirt is really something you might like after buying that other shirt. Humans are correcting customer-service chatbots, listening to Alexa requests, and categorizing the emotions of people on video calls. They are labeling food so that smart refrigerators don’t get confused by new packaging, checking automated security cameras before sounding alarms, and identifying corn for baffled autonomous tractors.

While open access is a cornerstone of Wikipedia’s design principles, some worry the unrestricted scraping of internet data allows AI companies like OpenAI to exploit the open web to create closed commercial datasets for their models. This is especially a problem if the Wikipedia content itself is AI-generated, creating a feedback loop of potentially biased information, if left unchecked.

It would be foolish to try to forbid Wikipedia contributors from using chatbots to help write articles: people would use them anyway, but would try to hide the fact. A ban would also be counterproductive. LLMs are simply tools, just like computers, and the real issue is not whether to use them, but how to use them properly.

While open access is a cornerstone of Wikipedia’s design principles, some worry the unrestricted scraping of internet data allows AI companies like OpenAI to exploit the open web to create closed commercial datasets for their models. This is especially a problem if the Wikipedia content itself is AI-generated, creating a feedback loop of potentially biased information, if left unchecked.

This analogy to lossy compression is not just a way to understand ChatGPT’s facility at repackaging information found on the Web by using different words. It’s also a way to understand the “hallucinations,” or nonsensical answers to factual questions, to which large language models such as ChatGPT are all too prone

When an image program is displaying a photo and has to reconstruct a pixel that was lost during the compression process, it looks at the nearby pixels and calculates the average. This is what ChatGPT does when it’s prompted to describe, say, losing a sock in the dryer using the style of the Declaration of Independence: it is taking two points in “lexical space” and generating the text that would occupy the location between them

they’ve discovered a “blur” tool for paragraphs instead of photos, and are having a blast playing with it.

A close examination of GPT-3’s incorrect answers suggests that it doesn’t carry the “1” when performing arithmetic. The Web certainly contains explanations of carrying the “1,” but GPT-3 isn’t able to incorporate those explanations. GPT-3’s statistical analysis of examples of arithmetic enables it to produce a superficial approximation of the real thing, but no more than that.

In human students, rote memorization isn’t an indicator of genuine learning, so ChatGPT’s inability to produce exact quotes from Web pages is precisely what makes us think that it has learned something. When we’re dealing with sequences of words, lossy compression looks smarter than lossless compression

Generally speaking, though, I’d say that anything that’s good for content mills is not good for people searching for information. The rise of this type of repackaging is what makes it harder for us to find what we’re looking for online right now; the more that text generated by large language models gets published on the Web, the more the Web becomes a blurrier version of itself.

Can large language models help humans with the creation of original writing? To answer that, we need to be specific about what we mean by that question. There is a genre of art known as Xerox art, or photocopy art, in which artists use the distinctive properties of photocopiers as creative tools. Something along those lines is surely possible with the photocopier that is ChatGPT, so, in that sense, the answer is yes

If students never have to write essays that we have all read before, they will never gain the skills needed to write something that we have never read.

Sometimes it’s only in the process of writing that you discover your original ideas.

Some might say that the output of large language models doesn’t look all that different from a human writer’s first draft, but, again, I think this is a superficial resemblance. Your first draft isn’t an unoriginal idea expressed clearly; it’s an original idea expressed poorly, and it is accompanied by your amorphous dissatisfaction, your awareness of the distance between what it says and what you want it to say. That’s what directs you during rewriting, and that’s one of the things lacking when you start with text generated by an A.I.

People are not interested in visiting websites about a topic; they, by and large, just want answers to their questions. Google has been strip-mining the web for years, leveraging its unique position as the world’s most popular website and its de facto directory to replace what made it great with what allows it to retain its dominance.

Artificial intelligence — or some simulation of it — really does make things better for searchers, and I bet it could reduce some tired search optimization tactics. But it comes at the cost of making us all into uncompensated producers for the benefit of trillion-dollar companies like Google and Microsoft.

Search optimization experts have spent years in an adversarial relationship with Google in an attempt to get their clients’ pages to the coveted first page of results, often through means which make results worse for searchers. Artificial intelligence is, it seems, a way out of this mess — but the compromise is that search engines get to take from everyone while giving nothing back. Google has been taking steps in this direction for years: its results page has been increasingly filled with ways of discouraging people from leaving its confines.

In other words, when Kylie Jenner posts a petition demanding that Meta “Make Instagram Instagram again”, the honest answer is that changing Instagram is the most Instagram-like behavior possible.

The first trend is the shift towards ever more immersive mediums. Facebook, for example, started with text but exploded with the addition of photos. Instagram started with photos and expanded into video. Gaming was the first to make this progression, and is well into the 3D era. The next step is full immersion — virtual reality — and while the format has yet to penetrate the mainstream this progression in mediums is perhaps the most obvious reason to be bullish about the possibility.

The second trend is the increase in artificial intelligence. I’m using the term colloquially to refer to the overall trend of computers getting smarter and more useful, even if those smarts are a function of simple algorithms, machine learning, or, perhaps someday, something approaching general intelligence.

The third trend is the change in interaction models from user-directed to computer-controlled. The first version of Facebook relied on users clicking on links to visit different profiles; the News Feed changed the interaction model to scrolling. Stories reduced that to tapping, and Reels/TikTok is about swiping. YouTube has gone further than anyone here: Autoplay simply plays the next video without any interaction required at all.