Meta has gone after AI in the same way they went after the metaverse, by splashing money around and rushing to build products fast. They’re spending tens of billions building out data centers and related AI infrastructure. They’re tossing out incredible compensation packages in the hundreds of millions of dollars to top AI researchers.

Facebook is now the cultural symbol for useless slop and disinformation, while ‘that’s so Reels’ is now a common insult for terrible shortform video content - and you get the first theory of Meta. It’s a laughable company whose core business is increasingly uncool, a company in decline, a company that falls flat on its face any time it tries to change things up.

Call Meta uncool all you’d like, metrics are up across the board. They’re getting higher user engagement, higher user counts, and they’re selling more ads at a higher price-per-ad. The numbers are up in virtually every way on every platform - Facebook, Instagram, WhatsApp and even Threads.

They can afford to make huge bets on speculative new technologies because they have more money than they know how to spend. Why not spend all that money on the metaverse or on AI? What else are they going to do with it? Zuckerberg still controls the company and he’d prefer to invest in new technology rather than just pay himself fat dividends. This is the second theory of Meta - a unbelievably successful company whose core business is booming and who spends a lot of money on speculative investments simply because they can.

Reinforcement learning with human feedback is a process by which models learn how to answer queries based on which responses users prefer most, and users mostly prefer flattery. More sophisticated users might balk at a bot that feels too sycophantic, but the mainstream seems to love it. Earlier this month, Meta was caught gaming a popular benchmark to exploit this phenomenon: one theory is that the company tuned the model to flatter the blind testers that encountered it so that it would rise higher on the leaderboard.

A series of recent, invisible updates to GPT-4o had spurred the model to go to extremes in complimenting users and affirming their behavior. It cheered on one user who claimed to have solved the trolley problem by diverting a train to save a toaster, at the expense of several animals; congratulated one person for no longer taking their prescribed medication; and overestimated users’ IQs by 40 or more points when asked.

OpenAI, Meta, and all the rest remain under the same pressures they were under before all this happened. When your users keep telling you to flatter them, how do you build the muscle to fight against their short-term interests?  One way is to understand that going too far will result in PR problems, as it has for varying degrees to both Meta (through the Chatbot Arena situation) and now OpenAI. Another is to understand that sycophancy trades against utility: a model that constantly tells you that you’re right is often going to fail at helping you, which might send you to a competitor. A third way is to build models that get better at understanding what kind of support users need, and dialing the flattery up or down depending on the situation and the risk it entails. (Am I having a bad day? Flatter me endlessly. Do I think I am Jesus reincarnate? Tell me to seek professional help.)

But while flattery does come with risk, the more worrisome issue is that we are training large language models to deceive us. By upvoting all their compliments, and giving a thumbs down to their criticisms, we are teaching LLMs to conceal their honest observations. This may make future, more powerful models harder to align to our values — or even to understand at all. And in the meantime, I expect that they will become addictive in ways that make the previous decade’s debate over “screentime” look minor in comparison. The financial incentives are now pushing hard in that direction. And the models are evolving accordingly.

Unlike the “trauma plot,” Parul Sehgal’s coinage for the use of trauma as narrative payoff, the therapeutic plot doesn’t wallow in trauma itself. Instead, it offers formulaic accounts of diagnosis and healing—what Janet Malcolm has called “the streamlined truisms of the age of mental health.”

Book blogs sort recommendations by pathology (severe social anxiety, schizophrenia, body dysmorphia), symptom (anxiety, panic attacks), and trauma (parental suicide, psychiatric stay). Fans diagnose characters with mental disorders (a fan theory diagnoses the character Bruno from another recent Pixar offering, Encanto, with obsessive compulsive disorder). Self-diagnosis even informed the development of Everything Everywhere. In early drafts of the script, Evelyn suffered from undiagnosed attention deficit hyperactivity disorder. Writing and researching her character inspired Kwan to identify and seek treatment for his own heretofore undiagnosed ADHD.

No doubt, such a medical diagnosis can provide relief and resolution. But is such a diagnosis the job of art? What is lost when audiences and creators eschew other ways of discussing fiction (for example, E.M. Forster’s distinction between round and flat characters) and instead reduce characters to clinical profiles?

In ancient Athenian tragedy, catharsis was defined by Aristotle’s Poetics as the ritual purification and purgation of emotions, particularly pity and fear. Pity arises from identification with the tragic hero, whose nobility is compromised by a fatal flaw, and fear is elicited by his excessive punishment. The therapeutic significance of catharsis originated much later in the theories of Sigmund Freud. By applying Joseph Breuer’s “cathartic method,” Freud theorized that hypnosis allowed patients to recall the traumatic experience at the root of their condition. Catharsis was Freud’s first major breakthrough, and his first brush with the powers of the unconscious that would form the underpinnings of psychoanalytic theory.For Aristotle, catharsis was the result of anagnorisis—the humility produced by the tragic hero’s recognition not only of the calamity that had befallen him, but also of his own role in bringing it about. Freud, meanwhile, described an inherent tragedy in the “impossible profession” of psychoanalysis, “in which one can be sure beforehand of achieving unsatisfying results.” It wasn’t that Freud had no faith in his own methods; he simply perceived the enormity of the human condition, and understood that the odds of success were not stacked in the psychoanalyst’s favor.Humility is absent from today’s therapeutic catharsis, which assumes with algorithmic certainty that sharing will lead to understanding, and that understanding will lead to healing. Art’s role, according to Everything Everywhere actress Stephanie Hsu, is “to hold space for trauma and offer catharsis,” and to recognize that “empathy and radical empathy and radical kindness are also a tool.” Buried beneath this gauzy language is the fact that the “empathy” of Turning Red and Everything Everywhere rely on the transformation of the mothers, not their children. The adults must learn that children are individuals rather than extensions of parental will, and when empathy is granted to mothers, it is only through their shared status as victims.

Pity the immigrant women who fled war-torn nations and corrupt regimes only to be subjected to psychoanalysis from hipster filmmakers and their own children

All this is not to say that storytelling holds no empathic power, nor an ability to transcend individual perspectives. But this power lies in art’s ability to overthrow, not reify, easy solutions—to challenge rather than “validate.” The transmission of suffering from one generation to the next is a worthy subject for art, but not because its effect on any particular demographic has been under-represented.

Our approach to culture should account for rigor and complexity, not defer to trite solutionism.