Is it an art-house short or a B2B SaaS ad?
Every AI launch looks like an A24 trailer, and Silicon Valley has decided its next competitive edge is taste. We set out to answer what taste really means—and if a machine can have it.
“Hey Siri, play ‘Fool’ by Perfume Genius.”
Her eyes linger on you for one second before darting down to focus on her taste buds. She twirls a forkful of pasta into her mouth, the fork gently clinking against her teeth before her lips close and form a smile full of pleasure. The little director in your mind punches the air in triumph.
“Good?” you ask.
“Mm…” she coos reassuringly as she finishes chewing. “Really good,” inflecting in slight surprise, following you to help set the table.
Your shoebox apartment is a flood of golden domesticity. Forget the fiddle-leaf fig clinging to dear life in the back window, or the books your roommate left everywhere—thank goodness you took two minutes to stack them, horizontally, on the console shelf. The paper lantern you bought this weekend in a flatpack from IKEA is doing its thing, humming its sweet summer nectarine glow. A faint echo of city rush-hour leaks through the old sash windows. You’re here—eating linguini with your amour, seeing it all play out in 35mm film, and you can’t help but think: this life is a movie.
Then words appear on the screen:
I need a recipe that says, “I like you, but want to play it cool.”
You realize that you’re not watching your first date with your future wife—you’re watching Dish with ChatGPT, an installment in OpenAI’s latest ad campaign. Director Miles Jay gets only twelve seconds of the half-minute runtime to establish the pretense of nervous chemistry on an early date.
A ChatGPT response scrolls upward—like movie credits—in clean, serif text: “Here’s the move: Lemon-Garlic Butter Pasta with Cherry Tomatoes.” The camera pulls back—out the window, across a brick façade glowing amber in the dusk—a drone outside the building. As the last sentence reads, “Above all, don’t sweat it. You got this,” just above ChatGPT’s logo, you become a voyeur of someone else’s intimacy. We leave the human story behind, left in the cold harsh air, as a voiceless wall of text takes the credits for the intimate production of our lives. If you’re feeling spooked, join the club.
AI companies are having their Mad Men moment, borrowing art-house film to sell feeling, whether warmth (ChatGPT) or resolve (Claude). It seems like every AI product launch now arrives with its own high-end short film on LinkedIn or X, each promising that this one will change everything. (I’ll gladly take this opportunity to plug Osmo, a creative studio I advise that actually makes these things.)
These commercials’ familiar aesthetics—with ChatGPT’s grain and intimate awkwardness akin to something out of A24—pull us into stories we’ve seen before, a kind of déjà vu where the story of our lives is not given new life but recycled from old cultural fabrics. But they’re not movies; they’re ads. And they’re not selling feeling; they’re selling the idea that AI models can understand our feelings—like first-date jitters—so intimately as to coordinate them alongside us. The irony is that the only thing AI about these commercials is the product they’re selling; these campaigns rely on humans—indie agencies and film-school directors—and even they, cherry-picked for their sense of taste, struggle to make the machine feel tasteful.
taste
But before we get into that, let’s have a proper conversation about taste.
I’m a foodie, and I love words, so I care a lot about getting this right. “Taste” has somehow become Silicon Valley’s latest blanket buzzword—we reach for it when we want to sound pro-human (next time, just reach for Artificial Whimsy). Wharton’s “AI, Taste, and the Future of Creativity” warns that creativity without taste risks becoming “beautifully meaningless.” The Atlantic calls it “the instinct that tells us not just what can be done, but what should be done.” And Julie Zhuo wonders when “AI might have better taste than you.”
Taste is the human act of discernment, the narrowing of infinite possibility into choice. This choice is not made in a vacuum: art is always in conversation, and to have taste is also to anticipate resonance—to know whom you’re talking to, what will move them, and why. It’s what turns information into intention, and ultimately into communication.
If creativity expands, taste edits.
Last week, I went to a talk by Alyosha Efros, a computer vision professor at Berkeley and one of the most vocal advocates for the primacy of data in AI. He describes modern AI as “data plus plus” (data++), not novel intelligence, but civilization compressed. With enough unlabeled visual data, he argues, models can move beyond rigid, human-made categories and let data reveal its own patterns, in a process known as self-supervised learning. In Efros’s framing, the question isn’t “What is this?”, but instead: “What is this like?”
The shift matters because it implies meaning doesn’t have to come from humans—it can come from data itself. Models are engines of convergence: they compress variation toward the statistically likely center of the data manifold.
creativity
I wonder if creativity has always been this: remembering beautifully. As Wordsworth famously wrote in his “Preface to the Lyrical Ballads,” “poetry is the spontaneous overflow of powerful feelings: it takes its origin from emotion recollected in tranquillity.”
Memory and creativity are often described as two distinct aspects of human cognition: one involves the retention and recollection of facts and experiences, while the other embraces the generation of new ideas. Still, the same brain networks that help us recall past experiences, such as the one active during daydreaming, support what psychologists call imaginative recall: the recombination of stored memories into new configurations. To compose a piece of music, we draw unconsciously on everything we’ve ever heard, from the twelve notes that underpin Western harmony to Bach’s counterpoint, Chopin’s phrasing. That, of course, is what AI does best, minus the intent.
Models essentially industrialize the act of remembering. They comb through fragments of our collective past at a scale no human mind could, but will always lack the emotional filter that tells us why something belongs in the composition. What we call creativity in modern AI is largely an illusion built on interpolation.
As Efros points out, modern AI isn’t discovering new ideas, but operating inside an enormous, high-dimensional space of human experience and drawing straight lines between known points. To quote Arthur C. Clarke, “Interpolation in sufficiently high-dimensional space is indistinguishable from magic.” The magic, of course, is a trick of scale: pass enough data through a simple algorithm, and it begins to look profound. What feels like originality is often the statistical midpoint of a thousand neighboring examples.
But there’s a ceiling to what interpolation can do. Since the 1990s, vision researchers have known that any new face can be represented as a linear combination of roughly two hundred training faces—there are, as Efros jokes, “more people on Earth than faces.” The same principle applies to creativity. When the raw material is culture, models can only recombine what’s already been seen.
The difference between remembering beautifully and remembering industrially lies in intent: humans imagine with purpose, machines with probability.
the next frontier
In traditional filmmaking, directors historically served as translators between concept and image, encoding meaning in each shot. Maneesh Agrawala, a Stanford professor who studies human–machine interfaces in creative work, names the central challenge of AI-assisted art: control. “Unpredictable black boxes are terrible interfaces,” he says, arguing that the real bottleneck in creative AI is in direction rather than generation. The more controllable a system becomes, the more expressive it can be.
We talk a lot about “generative” AI “revolutionizing” human creativity, but generation is only half of the equation; the next frontier of AI usability will not be its generative possibility, but rather its assistive possibility: its responsiveness to human vision, direction, revision—its responsiveness to human taste.
That artists can have assistants is no novel thing. Andy Warhol named his studio “the Factory,” using assistants to industrialize the realization of his creative direction. Centuries earlier, assistants helped Michelangelo paint the Sistine Chapel. AI may be the next creative assistant; it may even be the next creative interface. But a human still needs to imagine, envision, direct the art, if not also make it.
