Imagine interviewing people who silence their phones during dinner. Someone labels the comments “digital fatigue.” By the time an AI-generated summary reaches the client, the finding reads: “Consumers are rejecting connectivity.” The interviews may be accurately quoted, yet the conclusion says something the participants never said. A habit has acquired a motive, and a few people have come to stand for a market. To support that conclusion, the team would need to establish why people put their phones away and how widely those reasons are shared. If it also claims that attitudes are changing, it needs evidence from more than one point in time. In cultural research, drift happens when an explanation changes as the work passes between people and tools and nobody checks what justified the change.

Read more

Last week’s Labubu launch makes a useful research exercise: what might be causing the character’s decline, and what does the answer say about adults and play beyond toys? POP MART put the new series on sale with Labubu dressed as a cat, and resale prices split within a day. The secret figure traded at close to seven times the box price, while the cheapest listings for half the standard figures fell below it. Analysts said larger production runs had punctured the resale bubble, and some headlines have already declared Labubu dead. Many collectors objected to the design. It keeps Labubu’s pointed elf ears and stacks a pair of cat ears above them, and commenters on the official account called it odd and AI-stitched. POP MART’s US help page says a blind box opened online is “considered consumed, similar to a digital product.” Read together, the complaint and the policy describe two products in one box: a surprise that ends at the reveal, and a face owners know well enough to defend.

Read more

In a 1983 conversation for Channel 4’s Voices, John Berger told Susan Sontag about a shepherd whose life he had wanted to keep in a story. He began with what he knew of the man. Sontag cut in: “And then you started inventing.” Berger agreed, and held that the invention stayed tied to the truth of that life. The exchange appears in To Tell a Story, the collection of their conversations and letters that Canongate published this year, in Berger’s centenary year. Elsewhere in the same conversation Sontag says storytelling faces toward truth and toward invention and lies at once. She let both directions stand.

People who talk about artificial intelligence usually pull those directions apart. When a model invents a fact, we call it a hallucination. When it invents a plot or a character we like, we call it creativity. Both come from the same machinery. A language model generates text piece by piece from patterns learned in training and from the context it is given, and that process can produce an invented court opinion as readily as an invented ghost. In a single afternoon, a strategy team may ask the same model for a provocation to open a workshop and for figures to put in a client deck. The difference sits in the request. A ghost story asks the reader to enjoy it, and a legal citation asks the reader to rely on it.

Read more

On Thursday night a Tesla Roadster drove upside down along a curved ramp at SpaceX’s test site in McGregor, Texas, then fired cold-gas thrusters and flipped back onto its wheels. Musk wrote on X, “Turns out the future does have flying cars.” Tesla presented the car as “the best of the last human-driven cars,” yet the driver’s seat was empty, which Musk told the crowd was “to make sure we don’t kill anyone.” Tesla says the flying Rocket trim is not approved for public roads, and Musk puts production at 2028. The Roadster was first shown in 2017.

Read more

This April, six ventures led by students and recent alumni of the Maryland Institute College of Art presented their businesses to judges and an audience in Baltimore. They were competing for $105,000 in seed funding at the tenth annual UP/Start competition. The teams had spent four months developing their ventures before the pitch. One ceramics student hoped to take her studio global. The largest award went to an arts collective that stages exhibitions in the city’s underused spaces. The competition gives students money to continue their work. It also asks them to demonstrate something a studio critique does not: that the work can sustain an organization.

A venture panel decides whether a proposal deserves capital. A ceramics critique asks whether the form, surface and glaze work together. Students may need to answer both questions, but the answers are not interchangeable. A convincing business can sell unresolved work. Accomplished work can struggle to support its maker. An art school has to decide when each judgment belongs in the education.

Read more

Earlier this month OpenAI added a Try on button to ChatGPT’s shopping results. A shopper uploads a photo and ChatGPT generates a picture of that person wearing the product. OpenAI keeps the photo for later sessions. Google has offered a version of this since last year. Last December it began letting shoppers start from a selfie. The shopper uploads the selfie and can select a usual clothing size. The system returns several generated full-body images, and the shopper picks one as the default for later try-ons. Google’s help page offers the size field as a way to improve body shape accuracy. The face is drawn from a photograph. The body is estimated, and a clothing label can be one of the inputs.

Read more

A raised phone is a familiar cue that someone may be filming. The cue comes from handling the device, so a person filming that way gives it without choosing to. It is also approximate, since people hold phones up without recording and record without holding them up. Smart glasses remove the need to lift a camera. The wearer records with a press on the frame or a spoken command, and Meta puts a small white light beside the lens.

Read more

Cheap AI images have weakened the connection between a polished picture and the human work required to make it. Semioticians call a sign physically or causally connected to what it represents an index. A handmade image carries traces of its maker’s actions, but its appearance alone can no longer establish how it was produced. Brands have long displayed craftsmanship: Gucci brought artisans into its stores to demonstrate production in 2010. AI gives that practice another function: establishing that human work took place at all.

Read more

Sophie Kemp discovered that she belonged to a literary category by Googling herself. Her novel, Paradise Logic, had found readers on a Reddit forum devoted to ”weird girl literature.”

The community recommends books about women whose lives and desires depart from social expectations. Its categories include alienation, romantic delusion and body horror. For readers, the name provides a way to find books and other readers that more conventional classifications might miss. Some of the writers being recommended are less comfortable with it.

Read more

Read more

Read more

Read more

Read more

Read more

Meta’s new AI agent works behind a face. Muse can send email, make purchases, and keep working after the app is closed. Its default face is a small cream-colored cartoon creature called Jolly. According to Axios, Jolly appears at the top of the app tapping at a keyboard while Muse works. Its designer, Alex Cornell, counts that status display among his favorite parts of the design.

Meta’s largest rivals in consumer AI have generally used abstract marks. OpenAI represents voice activity with a swirling circle. Google built Gemini around a four-pointed sparkle and a visual language of circles and fluid shapes. These forms can indicate activity without implying a creature behind it. Meta chose a creature.

Read more

Information in a scene rarely arrived alone. A session drummer’s name in the liner notes led to another record. A clerk who knew why the two records belonged together sent you to a photocopied zine, and the zine sent you to a label and a late-night radio show. Each step also taught you which critic the clerk trusted and which shop carried the import. The sociologist Sarah Thornton called the status that came from this kind of knowledge subcultural capital, the standing of being in the know. In a record scene, much of that knowledge concerned how things connected. The same routes kept people out. A clerk could explain a record to one customer and sneer at the next, and a zine’s in-jokes marked who was already inside.

Related Reading: Crate Digging for Intelligence Analyzing Trends March 2025

Parts of the early web borrowed that structure and made it visible. Many hand-built fan sites linked outward, and each link showed what the maker thought belonged beside their own page. A blogroll made a writer’s chosen connections visible in the sidebar. Webrings went further. Sites around a shared interest joined a ring, and a “next site” button carried the reader from one member to the next, so arriving on a page placed you inside a circle of people who had agreed to be found together. Search reordered that field long before AI arrived, and optimization and advertising shaped what searchers saw. The ranked results page kept one property of the hand-made link. It showed a list of other places to go.

A generative answer often speaks in one voice, in a finished paragraph, with its sources gathered into small marks at the edge. Its form borrows the authority of the reference book. A list of results tells the reader that alternatives exist. A synthesized paragraph tells the reader that the comparison has already been made. The paragraph can be correct and still carry that message. A Pew Research Center browsing study from last year fits that reading: Google users clicked a traditional search link roughly half as often when an AI summary appeared. The study tracks where readers went next and says nothing about whether the summary was accurate.

The cost is easiest to see in material whose value a small group argued into existence. Record collectors have preserved histories around releases that barely sold, and fanzines have often championed bands before the larger music press paid attention. Klosterman gets at the stakes in the same interview. If a publication ever produced a list of the hundred greatest records and got it right, he says, it would be “the ultimate end of culture,” because culture would be “frozen to a point where we agree.” His book But What If We’re Wrong? rests on a related wager, that present certainty is provisional and later readers will rearrange it. Rearranging it depends on getting back to the material and to the arguments people made about it at the time.

Ask an AI about the experimental German music of the seventies, and it may hand you the word krautrock in seconds. The word works as a way in. British music writers helped popularize it, and many of the German musicians it grouped together disliked it and preferred other names for what they were making. An answer that supplies the term and leaves that dispute alone has handed the reader a password. There is a difference between receiving the password and entering the scene that made the password meaningful. Suppose these systems become better than search at surfacing the obscure record. That would be a gain. The record could still arrive without the argument that made anyone care about it.

Related Reading: Shibboleths in Branding Analyzing Trends, July 2025

A citation points to a source behind a sentence. Cultural provenance asks who connected two things, and against which reading they did it. An answer system could carry that kind of provenance by treating its sources as positions in an argument, with the people who disagreed kept beside the conclusion. Ask what krautrock is, and the answer arrives in seconds. Ask who named it and who objected, and you learn whether the route came with it.

Read more

Read more

Read more

Read more