Are you the crazy one?

There's one in every relationship.

You can listen to an audio recording of the text below, which might optimize its creepiness:

“Why is doing a reference check with three companies that employed someone you’re considering hiring a no-brainer, but talking with three exes of someone you’re considering dating is called stalking?” asks OpenAI CEO, Sam Altman, feigning indignation and incredulity. “Why is it normal for co-op boards to do exhaustive background checks but it isn’t for someone who’s beginning a romantic relationship?” Sam hasn’t blinked once in the several minutes he’s been peppering me with rhetorical questions. I wonder if perhaps he’s timed his blinks to coincide with mine to maximize eye contact. I also wonder if I’m actually talking with Sam Altman or with a prototype of a future OpenAI product. Sam is on a sort of press junket following the initial release of OpenAI’s latest product, which integrates the behavioral data feeds of all the major consumer technology companies to answer a question as old as sex: am I the crazy one? 

Every person in every romantic relationship spends half of their time in the relationship quietly scrutinizing their partner to determine the likelihood that this person will one day kill them in their sleep or worse, cheat on them with someone who lives in Hoboken. 

Unless they originate in Hoboken, relationships begin with blissful tranquility and lust and after six to nine months spiral into paranoid hellscapes because you notice a text message from an unfamiliar name associated with the gender they’re attracted to vibrate onto the preview screen of your partner’s phone very late at night while they’re peeing. You then google this name and discover that they are annoyingly attractive, and an Instagram search reveals that your partner follows and is followed by this person. This person, whose profile is private, has liked every single one of your partner’s grid posts for the past three months, which adds jet fuel to your sheepish insecurity about the fact that your partner has never posted or storied any photos of the two of you. 

For the next month you try very hard not to but totally end up side-eyeing your partner while they’re on their phone in bed next to you. When you ask a bit too cheerfully who they’re texting at 11:41pm on a Tuesday, they tell you, “my mom,” but even without your glasses you think you can see that there are more than three letters in the recipient’s name. On the evenings when you don’t have shared plans and they tell you that they have a work event, you contemplate catching up on your unread pile of New Yorkers on a stoop across the street from your partner’s home, but you decide that you’re not quite far enough out of your mind and instead take a gummy and binge Netflix. 

And that’s when you wonder – why hasn’t technology solved this problem?    

“One of our in-house baristas was telling me about the process he had to go through to buy his pied-à-terre in the West Village,” continues Sam, “surfacing every bit of his financial life, employment verification, interviews with everyone on the co-op board, who are all like 80-years-old and still have antennas on their TV sets – and I was like, here’s a twenty-five year old guy who makes almost a million dollars a year and has sixteen million dollars of stock in a prominent company, and he’s getting interrogated like a captured Russian spy in the 1970s just to buy some prewar two-bedroom apartment on the fringe of the West Village that he’ll maybe be in for a few weeks in Spring and Fall.” Sam and I both shake our heads for different reasons. “But like, he could walk out onto Folsom Street after work tomorrow afternoon, meet someone and marry them in a few months, and maybe - maybe he’ll stalk her Instagram profile for a few hours one night when she doesn’t text him back right away, or like, do an Internet search to see if she’s ever been arrested before he goes to Tiffany’s and spends a hundred grand on a diamond ring.” I wonder if this diamond is lab grown or mined. 

Like an inflatable venture capitalist, I press him to clearly articulate the problem he’s addressing with OpenAI’s mysterious new product. Sam narrows his eyes by about four percent, but he still doesn’t blink. A younger man walks into the conference room we’re in and sets an iced latte down on the table. I wonder if he’s worth sixteen million dollars and will be off to the West Village in late May. 

“Thank you,” says Sam to the young man. He takes a sip and returns his attention to me. “The problem is certainly the divorce rate,” says Sam, leaning forward a bit and uncrossing his legs. “But it’s bigger than that. Before divorce is even on the table, neither party in a relationship knows who the crazy one is, and there is always a crazy one. The key is to identify which person is the crazy one before marriage.” Sam leans back, nodding. “After organic produce from the farmers market, divorce is the second biggest expense for married couples in the state of California. The average divorce costs just under $4 million, and that’s not including virtual couples therapy with Esther Perel, which tacks on another $1.5 million.” 

“So, how does OpenAI solve this problem?” I ask, not bothering to question his figures, because I have divorced friends in Atherton and Mill Valley. Sam blinks once, but the blink is kind of like how Ron DeSantis smiles. 

“We’ve trained an LLM - a large language model - on consumer data provided by Apple, Google, Match Group, Meta, Microsoft, Netflix, Spotify, Uber, and Zillow, and without revealing any personal information the LLM gives the user a one-word answer that will empower them to make a decision that will save their romantic life.”

“So what’s the question?” I ask. 

“‘Am I the crazy one?’” says Sam. “Which is how we came up with the name of the product – AITCO, or aeyt-koh, we call it.”

“Cute,” I say. 

“And like, it’s got AI in the name, so-”

“Yeah, I get it,” I say. Sam waves his hand apologetically. “And the one word answer is what - yes or no?”  

“Yup,” says Sam. “I mean, correct. The one word answer is either yes or no,” he clarifies. 

“So, how do you know if the answer is accurate?” I ask. 

“That’s the easiest part,” he says with a smile. “Any one of the algorithms behind the social media, mobility and content tools we use can identify shady cheating liar behavior and paranoid weirdo behavior. Think about how Instagram knows who you’re dating but also shows you stories from people you want to sleep with, and how Uber remembers our home address and the other two addresses we’re always at, and how Spotify knows you want to hear sad songs when you’ve been dumped or when you create a sexy playlist and share it with someone, and how Hinge shows you people who look just like your ex. It comes down to this, right?” Sam pauses for effect. “AITCO analyzes fifty terabytes of historical data about a person’s location, their searches and social activity, what kind of content they consume, how that consumption coincides with their other activity, and their productivity at work. What you do on your phone is who you are, and when a user asks the four million dollar question, AITCO initiates a probabilistic investigation into both people’s behavioral data, and in less than four seconds our LLM knows who the crazy one is with 99.999 percent confidence. We’ve been in beta in the Marina in San Francisco for six months, and the product has been a huge success.” 

This strikes me as both dystopian and also very efficient, which is probably the best way to describe consumer Internet technology. 

“What does success look like?” I ask. 

“We’ve observed that when the user is told that their partner is the crazy one, the user terminates the relationship in an average of 51,408 minutes, which is about five weeks,” says Sam. “When the user is told that they are the crazy one, AITCO’s computer vision and object recognition coupled with Siri and natural language processing indicates that one in three users are engaged within a week and the rest remain together unless the other partner uses the product as well, in which case the relationship is terminated by the other user.”

“Savage,” I say. “Can a user see any, like, proof?” I ask.

“No, no, that would violate all kinds of privacy terms,” he says. “And, y’know, it would kind of take the fun out of it. Like, don’t you want to go down the rabbit hole and find stuff on your own and catch the person redheaded once you know you’re not the crazy one?” 

“Fair point,” I say. “So, straight up, Sam - this AITCO thing seems kind of far afield from OpenAI’s other endeavors, no?” I ask. “What does Satya think of this?”

“He’s a user,” says Sam, beaming. “He loves it.” 

“Is he the crazy one?” I ask. Sam smiles and waves a finger at me. 

“You know I can’t answer that,” he says. Sam glances at his $500,000 Greubel Forsey watch, slaps his thighs, and stands. “This has been fun, thanks for the great questions.” 

“Can I try the product?” I ask. “I’d love to be one of your New York City beta users.” 

“Actually we launched our New York beta two days ago, so yeah, I can do that,” he says, pulling out his iPhone. He taps the screen a few times and my phone vibrates. “There ya go. Have fun,” he says with a smile. 

I sit back down at the conference room table. An attractive younger woman walks by the open door looking at her phone. I wonder if she’s the crazy one in her relationship. I click the link that Sam has texted me. A plain screen reminiscent of ChatGPT opens, characteristically void of blustery consumer Internet product marketing and with only a few open fields for my and my partner’s full name, birthday, and Instagram handle. Centered below the empty fields is the acronym as a question, “AITCO?” on a grayed-out call-to-action button. My heart starts pounding. I wonder which is better – not knowing who the crazy one is, or knowing for certain. I decide it’s better to know. I fill in the empty fields, and the gray AITCO? button turns green. I click the button. 

One moment. I’m doing my homework, says the software in a chat bubble in the lower right of the screen. I count in my head. One. Two. Three. Four. 

And then the answer is on the screen. And it’s exactly what I thought. 

Thank you, CP, for being such thoughtful first eyes.