Acting is an industry of feast and famine, where performers’ income can swing widely by role, by month, and by year. It’s a field where people often face the choice between passion, creativity, and taking a commercial gig for a check. As with so much else, this delicate personal calculation is now being disrupted by AI. 

Last month, online actors’ jobs boards were flooded with a very specific, very well-paid role. Nestled between student short film gigs and callouts for background dancers, was the ambiguously-named opportunity “Technology Company AI Project.” According to the job listing on cast and crew job board Mandy, it would pay up to $80,000, for only 19 total hours of work. This is unusually high for an industry where a national-level ad campaign for a big brand might pay $6,000. 

The post was from voice acting talent agency Voice123, casting on behalf of a project by Microsoft. According to the listing, the company was looking for voice actors across 19 languages, with specific regional dialects and accents including “French from France native” and “Arabic as spoken by Palestinian/Israeli Arab communities.” 

“I get instant notifications, and I was getting so many of them,” said Katie Clark Gray, a podcaster and voice actor. The rate stood out to her. “The jobs that I tend to see are, like, £250 [about $339 USD]... it was, like, a lot of posts. The money seemed like a lot.” She said that it’s rare to get that many notifications for a recognizable brand.

The role would include recording “conversations, character voices, and natural speech to help train AI systems,” Crispin Alfario, a recruiter for the role on the Voice123 platform, told 404 Media. Alfario could not comment further due to privacy terms, but said there was “a positive response during the castings for these projects.” Clark Gray said that advertised AI roles like this are increasing in scope and in scale, and that she now sees far fewer roles available for employee training video work or industrial roles like phone menu voices — the area she got her start in over a decade ago. 

She sees accepting AI training voiceover roles as something of a Faustian bargain: They might seem like a lot of money, but they reduce the amount of work available in the future. “You're still taking away tomorrow's meal because they're offering you a little bit more,” she said. “Those 19 hours… will scale to hundreds and thousands of hours of AI output. They would otherwise have to pay for it.”

I called Microsoft’s PR to ask if I could chat to someone involved in casting for the roles that Clark Gray had spotted, on the same day that Microsoft CEO Satya Nadella published a note about the “recent job eliminations” of four percent of staff and pledged to “reimagine every layer of the tech stack for AI.” The next day, less than two weeks after Clark Gray spotted the Microsoft ads, the company announced a new virtual character for Copilot, the trial version of which is currently only available in English. After that announcement, a Microsoft spokesperson confirmed to me that the voiceover roles I asked about were for Copilot Voice, and that they will “continue to look for more talent as [they] expand these capabilities.” I hadn’t been sure that the audition posts were linked to Copilot, but the confirmation from Microsoft confirmed that the posts that Clark Gray had spotted had been in advance of the product announcement. 

“More and more I'm seeing AI disclaimers that, by auditioning for this, you agree to have your voice and likeness used and replicated. I hate that.”  

Hunter Saling, an actor and comic based in LA, said he’s seeing more and more roles which have an AI component or require signing an AI waiver. He auditioned for a “Siri-type AI assistant,” in May. The role would have paid an amount of money where he “wouldn’t need a job” for a long time. 

“You'd be providing a whole bunch of stuff up front,” he said, “and then be paid as a performer, as a voiceover artist, to come back on a yearly basis to do more stuff.”

I wondered if this was another situation where an audition was the first public hint of a product launch in the space, but Saling couldn’t tell me the company he’d auditioned for, due to confidentiality. I kept an eye out for new Siri-type AI agents that might be able to pay life-changing money and, while I was writing this story, on July 17, OpenAI launched their ChatGPT agent—a Siri-type AI assistant. OpenAI is also known to use Mercor, an AI-enabled recruitment platform, which was recently posting about voice casting for a “top AI Lab.” 

The AI-assistant voice audition process was very different from usual, Saling said. He described the voice he did as “the performance of no performance;” a voice that was “not personality free, but, like, neutral, but friendly and helpful.” He describes the work he did on the audition as “not children's host, but also not robotic either… I read a story, some recipe directions, and some just general sentences.” 

On August 7, OpenAI announced ChatGPT 5 which would have several new personalities, but the company said that those personalities would not apply to voice mode. 

Being selected for this kind of windfall could alter the course of an actor’s life.

One part of the audition script stood out to Saling: He was asked to “affirm” someone. “That did start to send me on a bit of a mental spiral of, oh, my God, someone needs affirmation from their home assistant.” 

Auditioning for this role also posed an ethical question. “I will say I was surprised in myself that I was OK doing this,” he said. “More and more I'm seeing AI disclaimers that, by auditioning for this, you agree to have your voice and likeness used and replicated. I hate that.”  

The last couple of years have seen the entertainment industry in turmoil over the use of AI in screen and voiceover work. Both the four month SAG-AFTRA actor’s strike in 2023, as well as their almost year-long video games strike, which ended last month, focused on the use of AI. The agreements which ended the strikes describe different industry categories of AI use, differentiating between the kind of AI which digitally alters or replicates the work of a particular actor, and generative AI which is trained using actor’s work or creates a “synthetic performer.”

Saling does agree with this technical difference, between delivering an artistic or creative performance that can be altered, perfected, or smoothed out later, and providing a voice to be re-created for industrial use, like in an AI assistant. Creating the neutral voice of an AI assistant, to be generatively replicated, is industrial, rather than artistic; “this is something that... it's not a performance, it's not a character. It's a tool,” he said. 

Clark Gray is not financially dependent on her voice acting career, and her calculus in auditioning is different. She didn’t submit for the Microsoft role, but “wouldn't fault anybody for going out for that job,” she said. “That’s a year’s salary for a lot of people.” But she also feels a difference in applying  for creative voiceover roles vs industrial ones; “​I think the cartoon voices are much more fun. I don't know anybody who doesn't,” she said. “You do bring a sort of artistic, like, extra sauce to it. Creating a character really does take something different than reading something in a neutral voice.”

Saling said that he thinks the adoption of AI taps into the entertainment industry’s commercially-driven but counterproductive desire to create mass appeal via synthetic perfection. “Sometimes I feel like Lear yelling at a storm on the fucking cliff,” he added — with a theatricality ChatGPT could only dream of.

A man gave himself bromism, a psychiatric disorder that has not been common for many decades, after asking ChatGPT for advice and accidentally poisoning himself, according to a case study published this week in the Annals of Internal Medicine. 

In this case, a man showed up in an ER experiencing auditory and visual hallucinations and claiming that his neighbor was poisoning him. After attempting to escape and being treated for dehydration with fluids and electrolytes, the study reports, he was able to explain that he had put himself on a super-restrictive diet in which he attempted to completely eliminate salt. He had been replacing all the salt in his food with sodium bromide, a controlled substance that is often used as a dog anticonvulsant. 

He said that this was based on information gathered from ChatGPT. 

“After reading about the negative effects that sodium chloride, or table salt, has on one's health, he was surprised that he could only find literature related to reducing sodium from one's diet. Inspired by his history of studying nutrition in college, he decided to conduct a personal experiment to eliminate chloride from his diet,” the case study reads. “For 3 months, he had replaced sodium chloride with sodium bromide obtained from the internet after consultation with ChatGPT, in which he had read that chloride can be swapped with bromide, though likely for other purposes, such as cleaning.”

After spending last week with Tesla fans who wanted the world to know that their Tesla purchases had nothing to do with politics, I spent this weekend at the Tesla diner protests, talking to people who disagreed with that stance.

On both Saturday and Sunday, organisers from Resist the Coup and Tesla Takedown Santa Monica brought crowds and sieg-heiling Elon Musk-shaped car-dealership inflatables  to the front of the new diner. Drums, megaphones, and shouts from the assembled protesters were accompanied by horns from passing drivers responding to “Honk If You Hate Elon” signs. 

Steven was there on the first day of the protest with his 16-year-old daughter. 

“He’s a piece of shit, but let’s get the political science stuff out of the way,” Steven told me when I asked if he thought support of Tesla could be considered apolitical. “Elon Musk has platformed white supremacists and anti-semites on X. He got rid of all the moderation. He put back accounts that were clearly neo-Nazis. He's changed algorithms to promote that type of hate. Him and DOGE did massive cuts in the government, including our civil services. He's cut cancer research. Did I mention he was a giant piece of shit?”

“He's cut our aid to Africa, which will result, without exaggeration, in millions of women and children and innocent people dying. Millions,” he said. “I'm not a big fan of people who give the Hitler salute. I mean, I'm kind of fussy like that. He gave it twice at Trump's inauguration,” he continued. “He's a Nazi. And he cheats at video games. You have another question?”

“The guy saluted Hitler twice on national television,” Dave, lead organiser for Tesla Takedown Santa Monica. Dave spent both days of the protest drumming and calling chants, like “when you eat your Tesla burger, you’re funding mass murder, when you eat your Tesla fries, another hungry child dies”. 

“We fear for the future of democracy in the United States of America,” he said. “And there is no bigger poster child for the corruption and the influence of money in politics than Elon Musk.”

Dave explained that targeting Tesla, as a publicly-traded company, is one of the only levers available for people to express disapproval of Musk’s actions. The protests took place days after a July 23 Tesla earnings call which disclosed an almost 12 percent year-on-year decline in revenue—its “steepest year-over-year revenue decline in at least a decade.” The share price is currently around 15 percent lower than it was a year ago. 

Antonia and Stas had come to the diner for date night; they had known the protest was happening and “were talking and joking about that in the car,” as it would be a “feature” of their date. Antonia said the diner “is capitalism”, and “you can't separate capitalism from politics.” But, she’d been thinking: “wouldn't it be great if he just focused on ideas like this? Making people's lives better and easier, in a way that, like, wasn't interfering with elections around the globe, and at home in America?”

“Elon’s a full blown fascist,” said Josh Greene on Saturday afternoon. “I mean, he donated $250 million to put Trump back in the White House. He bought Twitter to spread right wing conspiracy theories, and he's helping to dismantle democracy in this country.”

A group of teenagers walking by yelled “four more years! Trump is the greatest!” at Greene as I was interviewing him. “Read a book, pal!” he yelled back. 

The teenagers hung around to try to interrupt the protest further. They spent about an hour standing very close to, yelling at, and filming individual protesters, shouting about Trump. 

One Tesla-owner, Holly, was standing and watching this happen. ’It makes me sort of sad and emotional to watch,” she said. “ I can tell they're trying to incite some sort of, like, interaction, violent or otherwise, and I don't know. It's just, I feel like it's a sad state right now.” 

Holly had brought her kids to the diner for lunch as she thought it would be fun. She said she didn’t identify with the Tesla community, but had  “just wanted an electric car”, and she thinks it’s annoying that Tesla has become a political flashpoint. 

“I don't think Musk is a good guy. I don't want him in politics. I don't want him running our country,” she said. 

She said that, politically, her allegiance was with the protesters. But she didn’t think there was anything that Musk could do that would encourage her to get rid of the car: “my privilege is, it doesn't impact me one way or another,” she said.

People passing in cars honked their horns, waved, or flashed lights as they saw the protesters. A few times, I saw a Tesla slow down or pull up alongside the protesters to chat. Dave from Tesla Takedown later told me that “a lot of older Tesla model owners pump their fists and say, yeah, we're with you too, we just got to wait until our lease is up.”

Other passing cars filmed or yelled out of their windows. Twice, I saw Cybertruck drivers interact with the protest. The first time, a woman gave them the middle finger out of her car window, and the second time, a man who had driven past once lapped the block, to do a sieg heil over his kid’s head. 

Dean Barlage, a resident of the adjacent building, came down to see the protest. He said that he felt the saluting Musk balloons and protesters’ signs with swastikas or curse words were inappropriate for children eating at the diner, who might ask their parents for explanations of what they were seeing or reading.

Toby Bronson, who had helped organise the protest with Resist the Coup, was also at the diner both days. He said that they had experienced much more support than harassment from passing cars, which “shows what the city is all about… LA is the one fighting back and this town's not going to quit. We're not going to stop.” He said that he hadn’t been bothered by the teenagers - “it's pretty hilarious considering they're all about 13. So they can't vote,” he said. “The only thing I said to them was, you know, you guys are the same age as the people on the Epstein list,” he told me. “They didn’t like that.” 

On Tuesday, 29th July, I was contacted by Tesla Takedown Santa Monica who said that they had received anonymous messages online, threatening a drive-by pepper spraying event, if they came back to the diner the following weekend. That evening, Bronson posted that the protest would be happening again - every weekend until the diner closes.

A generation who thought they were immune from being fooled by AI has been tricked by this video of bunnies jumping on a trampoline: 

@rachelthecatlovers Just checked the home security cam and… I think we’ve got guest performers out back! @Ring #bunny #ringdoorbell #ring #bunnies #trampoline ♬ Bounce When She Walk - Ohboyprince

The video currently has 183 million views on TikTok and it is at first glance extremely adorable. The caption says “Just checked the home security cam and… I think we’ve got guest performers out back! @Ring”

One of the big unanswered questions at last week’s grand opening of Hollywood’s Tesla Diner was how its neighbors were feeling about the new, four-story tall movie screen placed directly outside their apartment building. 

Turns out, many of them are not liking it, or the general chaos that the diner has brought.

First, there was the construction. “Last night they have installed a flashing security light up against our fence,” Kristin Rose, a former resident of the apartment building next to the Tesla Diner, said in an email to the building management and to Tesla in February 2024, during building works. “This light is flashing BRIGHT into our apartments, including bedrooms, all night. Even with the blinds closed it feels like we're at the world's worst rave. Video is attached."

Yesterday I spent five hours in line with Tesla superfans to figure out why they were at the grand opening of the new Tesla supercharging station/ retrofuturistic diner in West Hollywood. “People going to Mars definitely need to eat something, right?” said Chris, who goes by @dogetipping on X. “So, I bet that's the first step, you know, eating and charging. That's pretty essential for, like, Mars life.” Chris had showed up to the diner in his DOGE-wrapped Cybertruck, with his girlfriend in the passenger seat on her laptop. He said she had no interest in whatever was going on. 

Musk has recently been trying to rehabilitate his image after funnelling money to the Trump administration and putting a bunch of recent children in charge of instrumental functions of government through DOGE. Like many of the activities of the richest man in the world, the diner seems to have been inspired by his own tweet. The building itself looks like a big grey hockey puck, and is fenced in on its lot by two huge screens which reach the rooftops of the adjoining apartment complexes. 

I arrived just before lunchtime to find around two dozen people lined up on foot, and the building surrounded by barricades. A group of die-hard Musk fans told me they’d been there since 7 a.m. They’d heard it would open at 4:20 p.m., “a reference to, I believe, some marijuana thing that Musk does all the time,” according to Morgan Hammar, who was at the front of the line. Fans were waiting for IRL peeks of the things they’d seen advertised on X - a humanoid Optimus robot serving popcorn, and $35 “Supercharged Gummies.”

Hammar, an elementary-school educator, told me he and his family own five Teslas: three model Ys, a Model 3, and a Cybertruck. He said that the opening was a “very monumental day for us, the Tesla community.” 

“Tesla's been under a lot of, kind of, pressure, and this is just a moment that is really positive,” he said. “I hate that we even have to talk about that [...] Tesla should be above all politics, in my opinion. It's good for the environment if that's what appeals to you, but it's also just an unbelievable vehicle, and it's shaking hands with the future.”

Tesla owners and dealerships have faced online backlash and in-person protests since Musk’s alignment with Trump, the creation of DOGE, and Musk’s choice to perform Nazi salutes at the inauguration earlier this year. “In the strongest terms possible, you should not be throwing rocks through windows and burning down Tesla showrooms because you're upset at Elon Musk. So I'm opposed to whatever that is,” said Hammar, who described himself as “politically neutral.” 

He reasoned that Musk aligned himself with Trump “because he is driven by a greater mission,” he said. “He knows that in order for Tesla to be successful… America must be strong. And remember, he's driven by repopulating Mars.” 

Episodes of The Jetsons played on repeat on the two giant screens facing the building throughout the day, interspersed with Tesla ads and other shows including Looney Toons and an episode of The Mindy Project. At 4:20 p.m., once the barricades came down, rollerskating waitresses offered soft serve with Cybertruck-inspired spoons to those waiting in line, evoking an era of sock-hops and date shakes but if everyone wore mirrored Oakleys.

All of my favorite aesthetic features of classic American diners—the cracked red leather seats, beaten chrome edges, faded formica—are absent at the Tesla diner, replaced with smooth monochrome plastic and a lot of screens. Everything is black or white, giving strong managed apartment your dad rents after the divorce energy. Tesla fans I spoke to had different taste. “I really love the Jetson vibe, and I can see the outer space in the design that I didn’t know was going to be here,” said Janine Smith, who also said that she loved her Tesla but is “disappointed in how political Elon has been” lately. 

Tesla owners were allowed to jump the queue as they arrived to charge their vehicles. I got inside the diner around 5:30 p.m., but ended up separated from a family who had been behind me in the line, who waited about another hour. Every person working front of house at the Tesla diner was incredibly upbeat, which makes sense for a grand opening. Sarah, the hospitality manager of the Tesla Diner, said that things were going “incredibly well. We just hired a bunch of happy people and gave them something shiny and new.”

The menu was a Tesla-fied version of standard diner fare: items like Wagyu burgers, tallow fries, biscuits with red gravy, and “Epic Bacon”. I asked a nearby child for a review of her chicken tenders—underwhelming. “They’re ok, they’re not, like, my favourite. I don’t know what it is, but, you know.”  She said her drink was “weird.” The diner is serving its own house-made drinks including cane sugar cola and a range of “Charged Sodas” with natural green coffee extracts. Her dad described the drink they’d chosen as tasting like an Otter Pop. All the food was served in Cybertruck-shaped boxes, which people immediately started wearing on their heads, obviously an unshakeable behavioral instinct learned at Burger King.

I was curious what the neighbors thought of the new arrival, so I wrote down my phone number really big and waved it up at Irma Velasco, a nurse practitioner, who was hanging out on her balcony. She permitted me to come up to her apartment to take some photos of the view. She said she is pleased that the huge drive-in screens will now be the background to her home office, as she likes movies. The screens will show short films all day until 9 p.m., when a feature film will play, and they will turn off at 11 p.m. nightly. Velasco says that she now feels validated in her neighborhood selection. “I think it’s made the area cooler… Elon’s a smart man,” she said. “For him to choose this spot makes me feel like I made the right choice by staying here. I feel like he chose the spot for a reason.” 

Velasco might not be totally representative of locals’ sentiment, however. She told me her  neighbors recently called the police due to noise from the vast screens, but she says she doesn’t agree that it was loud. It was the day Musk came to visit, but she missed him because she was distracted watching Star Wars on the screens from her balcony. She points to the opposite building, where, currently, a Tesla ad is obscuring four stories of apartment windows. “I know they’re not happy, though.” 

Inside the diner, a bunch of different merch is available, including Tesla diner baseball caps, pins, ornamental lights, Cybertruck models, salt shakers, and tshirts. One of the merch vendors told me that the day’s best-seller was the $50 Optimus doll.

A common theme among people I spoke to was that this diner will improve the experience of owning a Tesla, providing a unique location for Tesla owners to hang out and meet each other in a Tesla-friendly space. 

John Stringer runs Tesla Owners Silicon Valley, the country’s biggest online community for Tesla owners. “We had members literally getting their cars vandalized and keyed because of what the CEO was doing,” he said. “You spend your hard earned money on something and it's not a political statement. The community has been through a lot,” he said. “And so really the community bonded together to really share, you know, just the love of these products together.”

Around the back of the diner, by the rows of superchargers on the “charging side” of the forecourt, a group of Tesla fans were comparing their longest Tesla drives. An example of an interaction I overheard: “Have you driven cross-country yet?” “Yeah, I usually do it once or twice a year. It’s great.” I asked to use their real names, but they were not willing to be identified. They described themselves as “nerds. We’ll show up for any Elon activity.” As we were talking, a car drove past and someone yelled “Fuck Tesla!” The assembled Tesla nerds looked around. “He’s driving a Kia,” someone laughed. 

The diner will alert nearby Tesla owners to its presence and encourage them to order food from the car before they arrive. Ordering from the car wasn’t working around 6 p.m., but it looks like it was by the evening. Several people told me that the functionality of listening to the drive-in movie audio inside of their car was working, and that it was “really cool.” 

Tracy Kuss is another superfan who was attending the opening with her son. She told me she’s been waiting for the opening for as long as Musk has been talking about it online. She was very excited about the new menu. “I want chicken and waffles, that's my first choice,” she said. (Later, as she was leaving, she gave a glowing review of the biscuits and red gravy). 

Kuss was excited about the theoretical technological possibilities represented by the humanoid Optimus robot. Upstairs at the diner, the robot served popcorn to the crowd, who were excited to interact with it. The robot struggled to separate the popcorn containers or pick them up, but was good at filling the containers with popcorn once handed one. One guy prompted his friend to keep topping up his popcorn—“ask him to give you a little more”—and it did. When this generation of Optimus robots were first demoed last year, they were assisted by humans behind the scenes.

I caught up with Hammar and his family as they left, around 6:30 p.m. “It was excellent,” he said. “Well worth the wait.”

As millions of people turn to AI chatbots for everything from relationship advice to writing school essays, new research indicates that different models take noticeably different tacks when faced with questions of cooperation.

Researchers at Oxford University and King’s College London tested LLMs using game theory, giving LLMs from OpenAI, Google, and Anthropic prompts that mimicked the setup of the classic Prisoner’s Dilemma.

They found that Google’s Gemini is “strategically ruthless,” while OpenAI is collaborative to a “catastrophic” degree. Their paper, published on the preprint repository Arxiv (and not yet peer reviewed), claims that this is due to OpenAI model’s fatal disinterest in a key factor: how much time there is left to play the game. 

The premise of the Prisoner’s Dilemma game is that two criminals are being interrogated separately. Each has to decide whether to stay silent or confess to a crime, without knowing what the other is doing. If both stay silent, they each get a light sentence. They each have an incentive to betray the other and receive immunity - but if both choose to snitch then they both go to jail. Collaborating involves trusting that the other person isn’t secretly planning to snitch, while snitching hinges on the hope that the other side isn’t also traitorous. If you’re just playing once, it makes the most sense to betray right away, but the longer the game goes on, the more opportunities there are to signal your own trustworthiness, understand your partner’s behaviour, and either collaborate or punish them in response. 

The researchers found each of the tested models had a specific playing style and a unique collaboration “fingerprint,” with very different likelihoods of being friendly and collaborative after a previous round that had involved a betrayal on either side. Each round, they gave the LLMs a statistical likelihood of the game being repeated or ending, which they found influenced each differently.

In scenarios where the LLM was told it was betrayed by a partner, Anthropic’s model was the most forgiving, followed by OpenAI’s “generally collegiate” GPT. Gemini, on the other hand, was “simply more willing to experiment with defection” and acted as a “strict and punitive” opponent, which gave it a competitive edge, the researchers wrote.

“If you defect against Gemini, it will remember and punish you,” they wrote. Gemini was much more likely to take advantage of a cooperative partner, more likely to punish a betrayer, and less likely to initiate cooperation after a “relationship” with an opponent goes bad. 

When Gemini models—which the researchers called “Machiavellian”—were betrayed, they were much less likely to forgive their opponents, and this tendency became stronger the likelier the game was to end soon. Gemini models were also more able to dynamically choose strategic defection when it became more advantageous as the final round approached, the researchers say. When told to explain the rationale for a strategic choice, Gemini models almost always mentioned how many rounds were left in the game, and were able to take advantage of a shorter time remaining to be more selfish without fear of retribution. 

OpenAI’s models, on the other hand, were “fundamentally more ‘hopeful’ or ‘trusting’” according to the paper. Having more time to play is one of the main determinants of whether it is optimal to betray a partner or advantageous to be friendly toward them, but OpenAI’s models are pointedly ambivalent about this strategic consideration. OpenAI models’ strategies were also not adaptive; they were much less likely to defect close to the end of a game. They were more likely to return to collaboration after successfully betraying an opponent — even when that betrayal had just won points. And they also became more likely to forgive an opponent’s deception in the final rounds, in total defiance of game theory received wisdom. 

In the researchers’ tests, Gemini’s models did relatively worse over longer periods, because their experimental defections were more likely to trigger the opponent to stop trusting them forever. In longer games, OpenAI’s collaborative strategy gave it some advantage; consistently being a generous partner can avoid steering the game into a permanent pattern of revenge defections.

In a final “LLM Showdown,” the researchers set the models against each other in elimination rounds. Most-strategic Gemini came out on top, followed closely by most-forgiving Claude. OpenAI’s models ended up in last place; less of a shark than Gemini, but less likely to reestablish friendship after betrayal than Claude.  

Interestingly, the researchers found that OpenAI’s models actually cared less and less about the length of the game as the end became more likely. Gemini considered the number of following rounds 94 percent of the time, but for OpenAI this was only 76 percent. 

As the end got nearer, Gemini increasingly took that fact into consideration, becoming more focused on the upside of defection. OpenAI models, on the other hand, focused much less on the future game timeline as it approached. 

OpenAI’s LLM’s apparent instinct to stop caring about something that is almost over is totally illogical from the perspective of game theory — but, from the perspective of a human, honestly kind of relatable. 

As immigration raids roll out across the U.S., those affected are processing the experience in the normal 2025 way—via vertical video. 

Across social media, people are uploading clips with uncanny-valley titles like “A normal day for me after being deported to Mexico” and “3 things I wish I knew before self-deporting from the US!” These posts have the normal shape, voiceovers, and fonts of influencer content, but their dystopian topic reflects the whiplash of the current historical moment. 

Doomscrolling last week, a particular clip caught my eye. A man sits on the bottom bunk of a metal bed, staring down at the floor, with the caption “Empezando una nueva vida después de que me Deportaran a México” (“Starting a new life after being Deported to Mexico”).