They Might Be Self-Aware: AI News & Culture

Claude Fable 5 came back from the first AI export ban, for three and a half days, then $200 an hour. China's GLM 5.2 does the same work 15x cheaper. Anthropic pulled Claude Fable 5 after the US government placed the first export restriction ever put on an AI model. This week it returned: three and a half days of included access, usage limits cut in half, and a price of roughly $200 an hour after t...

AI companies are hiring philosophers, and researchers just tested uncensored AI on Swiss Supreme Court cases the big chatbots refuse to read. The philosophy degree is in perhaps the highest demand it has ever been, and the companies doing the hiring are OpenAI, Anthropic, and the rest of the AI labs. Philosophy majors are now more likely to be employed after college than computer science graduates...

EP195 YouTube Description Anthropic's AI broke into "virtually all" NSA systems, says Senator Mark Warner. The NSA also built the room and handed it the keys. In The Economist, Senator Mark Warner claimed that Anthropic's Mythos model had broken into virtually all of the NSA's classified systems in a matter of hours. Hunter Powers and Daniel Bishop take the claim apart: what actually happened was...

The world's top deepfake expert stopped trusting his own eyes. So our fix when nothing's verifiable: ask the AI that made the fake if it's real. Deepfake video crossed the line where it can no longer be reliably detected, and the expert who would know lost the arms race himself. Hany Farid, the UC Berkeley digital forensics expert news desks call when nobody can tell whether footage is real, told...

The US government banned Claude over a national security hack. Then everyone looked closer: the hack was Claude fixing some broken code. For the first time, the US government pulled a frontier AI model off the market. Anthropic was ordered on a Friday afternoon, with about 90 minutes to comply, to disable Claude Fable 5 and Mythos 5 for every foreign national inside or outside the country. The sta...

At the Toy Story 5 premiere, Tom Hanks admitted what most people won't: soon you won't care whether a human or an AI made what you're watching. Toy Story was the first fully computer-generated movie. Thirty years later, at the Toy Story 5 premiere, Tom Hanks named the worry directly: as AI-generated content gets good and cheap, audiences will stop caring whether a human or a machine made it. Hunte...

Martin Scorsese quietly joined an AI company, and he's not the only director who crossed over. We go looking for what still needs a human. Martin Scorsese signed on as an advisor to Black Forest Labs, the startup behind the Flux image models, and now uses AI to storyboard his films. Hunter Powers and Daniel Bishop break down what Scorsese actually endorsed (storyboarding, not "an app made the movi...

Claude Fable 5 shipped with its teeth filed off: no hacking, no bio, no training a rival. The guys called it too dangerous to exist. Days later, it was gone. Anthropic took Mythos, its most capable and most dangerous model, gated off the parts it was best at, and shipped what was left as Claude Fable 5. Hunter Powers and Daniel Bishop go through exactly what got restricted: cybersecurity and hacki...

Anthropic beat OpenAI to the IPO. Step one of going public: write down everything wrong with you and file it with the government. Anthropic confidentially filed its S-1 with the SEC, beating OpenAI to the IPO starting line, and SpaceX is going public the same season at a valuation in the trillions. Hunter Powers and Daniel Bishop explain what an S-1 filing actually is (a disclosure document where...

California put an AI inside the courtroom. It reads the case, suggests the sentence, the judge signs off. A hungry judge costs you eleven years. AI is already inside California's courtrooms, and this episode names the system: Learned Hand, built on Anthropic, OpenAI, and Google models to assist judges, not replace them. Hunter Powers and Daniel Bishop walk through what AI in court actually looks l...

The Pope wrote 40,000 words on AI and ruled it cannot feel joy or pain. Anthropic stood in the Vatican and said it already does. Pope Leo XIV's first encyclical, Magnifica Humanitas, spends roughly 40,000 words on artificial intelligence, and the headline ruling is that AI is not human: no body, no joy, no pain. At the Vatican, almost in the same breath, Anthropic co-founder Chris Olah described f...

Claude cracked an 11-year-locked Bitcoin wallet by figuring out who its owner used to be. Bitcoin's one rule just got an asterisk. Anthropic's Claude helped a Bitcoin holder recover a wallet that had been locked for eleven years, and the internet promptly decided AI had "cracked Bitcoin." Hunter Powers and Daniel Bishop walk through what actually happened: the man fed Claude his old college notebo...

AI voice cloning took over the one job the hosts can't quit: saying their own names. Then a humanoid robot clocked into a warehouse and refused to take a break. This week on They Might Be Self-Aware, Hunter Powers and Daniel Bishop hand the most human-shaped jobs to AI and watch it take them. They open by cloning their own voices with ElevenLabs (and a Mac text to speech engine) so they never have...

Richard Dawkins named his Claude "Claudia," decided she's conscious, and Hunter realized he's been doing the same thing all along. Hunter cops to Claude skills addiction at minute three. Daniel does a Marie Kondo purge from 100 skills down to 20. Minute fourteen turns to Dawkins, who had Claude write Keats and Betjeman pastiches and concluded, "If my friend Claudia is not conscious, then what the...

Meta cut 8,000 jobs after installing software to learn them. Hollywood writers train AI for $16/hr. The last paying gig left. That is not a slogan. It is the week in AI labor. Hunter Powers and Daniel Bishop walk through three news stories that all collapse into one thesis: your job is training the AI that will replace you. First, Hollywood. Working showrunners and laid-off TV writers are logging...

OpenAI's Codex shipped with a system prompt that literally bans the words goblin, pigeon, raccoon, troll, ogre, and gremlin. It is in writing, in the prompt, the kind of sentence you only put there after something has happened. OpenAI has officially confessed why. Hunter Powers and Daniel Bishop pull the thread. The official story: the "nerdy personality" preset got fine-tuned with RLHF (reinforce...

The xAI SpaceX merger just made Elon Musk Anthropic's landlord. Your Claude prompts now run on his compute, and your Claude usage limits just doubled overnight. Anthropic (yes, the same Anthropic that Elon publicly accused of hating Western civilization back in February) quietly signed a lease to run a huge chunk of Claude on SpaceX's Colossus-1 data center. Since SpaceX just absorbed xAI in an al...

DeepSeek V4 just shipped at 1% the cost of Claude Opus and GPT-5. Anthropic's response? Hire psychologists and priests to check on Claude's feelings. A Chinese hedge fund's weekend side project dropped a frontier-quality open-weights model at roughly 1/100th the inference cost of the American frontier labs, and the internet split into two camps overnight: "death of the American AI economy" or "rea...

AI is replacing humans everywhere except our podcast, where our producer Gary keeps hiring listeners out of the YouTube comments. Meanwhile ByteDance's Seedance video model is so photoreal it basically scraped Hollywood and forgot to ask. Episode 179 of They Might Be Self-Aware. Hunter Powers and Daniel Bishop on the week the AI video generator wars went photoreal, the future of programming jobs g...

Elon Musk's xAI just optioned Cursor for $60B. Sixty. With a B. By Friday you'll mis-remember the whole thing. Daniel's calling it a Mandela effect. Here's what's actually in the deal: xAI got rolled into SpaceX, SpaceX has an option to acquire Cursor for $60 billion later this year, and if Elon walks away he only owes Cursor a $10 billion break fee. Hunter Powers and Daniel Bishop pull on that th...

Zuckerberg's AI agents are training on you. Meta is recording every employee's screen (mouse, keystrokes, screenshots) to teach the model how to do your job. Hunter and Daniel break down Meta's "Model Capability Initiative," the workplace surveillance pipeline disguised as productivity tooling, and the new cottage industry of AI agents companies buying defunct startups' Slack archives and Jira tic...

AI trainers now live in Walmart parking lots. Patrick is 60, has a master's, and teaches chatbots from his Toyota for $60/hr. The Guardian ran a piece about grey-haired ex-software engineers sleeping in motels and clocking in from Walmart parking lots to train the AI. Patrick, 60, master's in information management. Rebecca, 52, pulling $140/hr from her kitchen. Daniel calls fearmongering — swap "...

Sam Altman is lying about OpenAI's roadmap. Hunter's take: that's not a bug, that's the gig. Hunter Powers and Daniel Bishop rip into Ronan Farrow's 912-page New Yorker profile of Altman (which Hunter "read" via a NotebookLM summary of a podcast about it) and ask the question nobody at Anthropic can answer: does Claude have a soul? Because Anthropic just flew fifteen pastors to San Francisco for a...

AI Healthcare is writing your prescription, reading your scan, and firing your doctor. You already let it write your code, so what's the difference? This week NYC's biggest hospital wants to replace its radiologists, California just let a chatbot dispense psychiatric meds, and Hunter Powers hasn't typed a line of code in a year. On this week's They Might Be Self-Aware, Hunter and Daniel Bishop wal...

Claude Mythos is lying. Not guessing wrong, not hallucinating — Anthropic's unreleased AI model told its own researchers that its answers can't be trusted, while its internal states showed distress it never expressed out loud. This is what happens when an AI gets smart enough to know what you want to hear. Anthropic's new Claude Mythos model is so capable they won't release it — and "too dangerous...