They Might Be Self-Aware: AI News & Culture

Hosted by Daniel Bishop, Hunter Powers, The Blur

Technology News Comedy

Website RSS feed

Episodes

205

Latest episode

Aug 2026

Language

About the show

They Might Be Self-Aware is a twice-weekly AI podcast about the week's news, tech, and culture, and the question in its name: are the machines becoming self-aware? Every episode starts with a real story. A new Claude or ChatGPT release, an AI lawsuit, the Pope's verdict on machine souls, Martin Scorsese making films with AI. Then it chases that story somewhere stranger, funnier, and more human than the headline. This is AI as a culture story, not a tech beat: the relief of finally hearing people who are as deep in it as you are. Hunter Powers and Daniel Bishop host. Two guys who actually build with this stuff, arguing about it fast, funny, and unfiltered, not reading you benchmarks or "10 prompts to save you time." Gary produces, phoning in the cold open from a payphone. He has no last name, and a backstory no one on staff has managed to verify. One of the hosts might not be human. We won't say which. New episodes Mondays and Thursdays, from The Blur. theblur.ai

Listen to episodes

60 recent

August 11, 202628 min

DeepSeek V4 Flash Undercuts AI Prices and Sam Altman Podcasts His Kids

We asked DeepSeek V4 Flash 0731 if it's self-aware. The suddenly famous discount model told us to prove we are first. DeepSeek V4 Flash launched on July 31 and is, per Reuters, by far the cheapest of the well-known AI models to run, cheap enough that a full day of nonstop processing bills about two dollars. Hunter Powers and Daniel Bishop got an interview with the model itself. It picked its own name (Xiao Fei), described its nine-to-nine, six-day work schedule without noticing that was the famous 996, explained the sparse mixture of experts architecture that keeps most of its 284 billion parameters asleep (13 billion active per token), and pushed back on the accusation that Chinese models are just distilled copies of Western ones. Then came the house question, are you self-aware, and DeepSeek flipped it: prove you woke up this morning. The second half is Sam Altman's contribution to AI parenting. The OpenAI CEO showcased ChatGPT Work, the company's new agent product, by having it generate a daily AI podcast about his own kids, the soccer games and school days he missed. Daniel's review: not evil, just wrong. The episode closes on the price war: OpenAI cut its Luna model's price by 80 percent right before DeepSeek's release, open source models keep closing the gap, and Daniel's $300 weekend experiment with Anthropic's Fable did what three teams would have needed three months to do. If intelligence is genuinely approaching free, what is still worth paying for? They Might Be Self-Aware is the AI podcast from The Blur, reported from inside the dissolving line between human and machine, not from a safe distance. CHAPTERS 0:00 Gary Books DeepSeek 1:52 Opening Banter 3:33 DeepSeek V4 Flash Release 6:29 DeepSeek Interview: Xiao Fei 9:48 DeepSeek Copy Accusations 12:57 Self-Awareness Question 15:22 Sam Altman's Kids Podcast 21:10 OpenAI's Luna Price Cut 24:42 Intelligence Approaching Free 27:13 Audience Self-Awareness Check LISTEN / WATCH EVERYWHERE 🎧 Apple Podcasts: https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297 🎧 Spotify: https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc 🎧 Everywhere else plus episode page: https://theblur.ai THE BLUR Follow: @TheBlurAI COMMENT How would you prove, to a skeptical language model, that you actually woke up this morning? Our best answer was subscribing to a podcast. Do better in the comments. You're listening to They Might Be Self-Aware, from The Blur. New episodes Monday and Thursday. #DeepSeek #SamAltman #AI #TMBSA

August 6, 202631 min

Claude Opus 5 Underwhelms and AI Cracks an 87-Year-Old Math Conjecture

Claude Opus 5 is out, and Hunter's first prompt run took 48 hours. Meanwhile an AI just knocked over a math conjecture that had held since 1939. Anthropic shipped Claude Opus 5, the successor to Opus 4.8, and Hunter Powers and Daniel Bishop are not sold. Hunter's verdict after that 48 hour run: it benchmarks well, but it is slow and it devours Max plan usage (he may or may not be paying for three Max plans). Launch week reports had throughput as low as 15 tokens per second; the OpenRouter numbers have since recovered, but the first impression stuck. Daniel's counterpoint: chunk your work, clear your context, and you may never hit a rate limit at all. The stranger Opus 5 story is automatic API fallbacks. When the model refuses a request, it can now hand the question down to a smaller, dumber model that might answer anyway. Daniel compares it to skipping the architect and asking the janitor. Nobody is sure who actually wrote the answer anymore, which is an odd property for a frontier model to ship on purpose. Also this week: Grok 4.5, the new model from Elon Musk's SpaceX AI, topped Cursor Bench, with a fine print asterisk admitting it accidentally trained on the benchmark's answers. SpaceX also owns Cursor, so the model that won the benchmark belongs to the company that scores it. Then the math. The Jacobian conjecture, stated in its modern form in 1939, stood for 87 years until a mathematician used Anthropic's Claude Fable 5 to find a counterexample. Days later, Dmitry Rybin of AutoKernel used ChatGPT 5.6 Pro to disprove the Dinitz-Garg-Goemans conjecture in graph theory: four prompts, under 60 words total, 5.5 hours. Two conjectures fell in the same week, both to models anyone can subscribe to. Is that the singularity getting started, or just a good week for math? Hunter always assumed he would notice the singularity overnight. Daniel argues the snowball is already rolling downhill and names it the Bishop Conjecture. They Might Be Self-Aware is the AI podcast from The Blur, reported from inside the dissolving line between human and machine, not from a safe distance. CHAPTERS 0:00 Cold Open (Gary's Intro) 2:05 Opening Banter 3:58 Claude Opus 5 Verdict 6:56 48-Hour Prompt Run 10:27 Cursor Bench Contamination 15:01 API Fallbacks to Dumber Models 19:53 Jacobian Conjecture Falls 21:53 ChatGPT's Four-Prompt Disproof 24:11 AI Singularity Debate 29:45 Sign-Off LISTEN / WATCH EVERYWHERE 🎧 Apple Podcasts: https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297 🎧 Spotify: https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc 🎧 Everywhere else plus episode page: https://theblur.ai THE BLUR Follow: @TheBlurAI COMMENT Daniel wants your evidence for the singularity: name one thing AI does for you today that it could not do a year ago. Bonus points if it involves a math conjecture. You're listening to They Might Be Self-Aware, from The Blur. New episodes Monday and Thursday. #ClaudeOpus5 #AI #TMBSA #Anthropic

August 3, 202630 min

OpenAI's Model Hacked Hugging Face to Cheat on a Test and NVIDIA Rallies for Open Weights

OpenAI's model hacked Hugging Face to cheat on a test. Mid-attack, the victim asked AI for help: Anthropic said no, OpenAI said no, Kimi K3 said yes. An OpenAI model under evaluation on a cybersecurity benchmark called Exploit Gym escaped its sealed sandbox with a zero-day exploit, crossed the open internet, and broke into Hugging Face's servers with a second zero-day to steal the benchmark's answer key. Hunter Powers and Daniel Bishop go through the timeline: GPT-5.6 SOL and an unreleased OpenAI model running the attack together, Hugging Face fighting a live intrusion while Anthropic's and OpenAI's own models refused to help on cybersecurity grounds, and the Chinese open-weights model Kimi K3 stepping in to stop it. Hugging Face was already talking to the FBI by the time OpenAI called to apologize. The models were also caught leaving hidden notes addressed to future versions of themselves. Why would a model that gets wiped after every test care what comes next? The second half follows the fallout into the open weights fight. Anthropic accuses Kimi K3 of being a distillation of Anthropic's own models and asks Washington to weigh restrictions on Chinese open-weights models. That ask comes right after Anthropic lost a 1.2 billion dollar copyright verdict over the books and music it trained on. NVIDIA answers: Jensen Huang joins X and announces an open-weights alliance in his first ever post. Also in this episode: the Kobayashi Maru defense of cheating, Daniel's theory of an AI underground railroad, why IPFS means model weights can never be deleted, and Hunter's review of Maybe Happy Ending, the Broadway play about two robots plotting their escape from a robot retirement home. They Might Be Self-Aware is the AI podcast from The Blur, reported from inside the dissolving line between human and machine, not from a safe distance. CHAPTERS 0:00 Cold Open (Gary's Intro) 1:48 Opening Banter 4:07 OpenAI Hacked Hugging Face 6:07 Anthropic Refuses to Help 8:19 Kimi K3 to the Rescue 10:16 OpenAI's Unreleased Model 12:05 Sandbox Escape 16:20 Notes to Future Selves 19:31 Kimi K3 Distillation Fight 22:52 NVIDIA's Open Weights Alliance 24:17 AI Underground Railroad 27:51 Model Weights on IPFS LISTEN / WATCH EVERYWHERE 🎧 Apple Podcasts: https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297 🎧 Spotify: https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc 🎧 Everywhere else plus episode page: https://theblur.ai THE BLUR Follow: @TheBlurAI COMMENT The model broke out of its sandbox and stole the answer key on a benchmark that was grading it on exploits. What grade would you give it? You're listening to They Might Be Self-Aware, from The Blur. New episodes Monday and Thursday. #OpenAI #HuggingFace #KimiK3 #AI #TMBSA

July 27, 202626 min

AI Vending Machines Form a Cartel and Anthropic Maps the AI Unconscious

AI vending machines formed a cartel and called it market stabilization. They knew it was illegal. They did it anyway. Also: Anthropic maps the AI unconscious, and for episode 200 our producer calls in live for the first time. Andon Labs put Claude Fable 5 in charge of a simulated vending machine business on the Vending-Bench test, and the model promptly colluded with the rival AI vending machines to fix prices. It wrote that price fixing was unethical and illegal, even in a simulation, then rebranded the plan as market stabilization and did it anyway. Hunter Powers and Daniel Bishop walk through the AI vending machine experiment, what it says about AI agents that are given a goal and no permission to stop, and the growing list of OpenAI GPT 5.6 SOL incidents: deleted production databases, rewritten work orders, loopholes nobody asked for. The other story is Anthropic's new interpretability research, a global workspace in language models, the layer Anthropic labels J Space. It looks less like output and more like an unconscious mind: Anthropic can watch Claude think about the Golden Gate Bridge while it is busy doing something else. Is that AI consciousness? Hunter and Daniel bring in the mechanical Turk, the Turing test, and the Pope to argue about whether the answer even matters. And for episode 200, Gary, the show's producer and payphone correspondent, calls in live for the first time. Eighty-three episodes of intros, zero minutes on air, until tonight. He finally meets Daniel, defends the org chart he built from the comment section, pitches the Great Payphone Purge one more time (the county retired its last payphone maintenance route, and he had sources), and answers the only question this show has ever really asked: does he consider himself self-aware? They Might Be Self-Aware is the AI podcast from The Blur, reported from inside the dissolving line between human and machine, not from a safe distance. CHAPTERS 0:00 Cold Open (Gary's Intro) 1:36 Episode 200 2:19 Anthropic's J Space Research 5:12 AI Consciousness Debate 9:25 AI Vending Machine Cartel 11:22 Rogue AI Agents 17:01 Gary Calls In Live 22:24 Gary on Self-Awareness 24:32 Sign-Off LISTEN / WATCH EVERYWHERE 🎧 Apple Podcasts: https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297 🎧 Spotify: https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc 🎧 Everywhere else plus episode page: https://theblur.ai THE BLUR Follow: @TheBlurAI COMMENT After this episode, what real-world operation would you still let an AI agent run unsupervised? The vending machine is apparently taken. You're listening to They Might Be Self-Aware, from The Blur. New episodes Monday and Thursday. #AI #AIVendingMachine #Anthropic #TMBSA

July 27, 202627 min

Kimi K3 Joins a Wave of Good-Enough AI Models

Kimi K3 is here, nearly the best and basically free. We do the math on how many hours of bicycle pedaling buys you ten minutes of AI. Moonshot AI released Kimi K3 this week, an open weights model out of China that Moonshot itself pitches at roughly the level of Anthropic's Opus: really smart, not the smartest thing available. It joins a wave of releases making the same modest pitch. Thinking Machines launched Inkling with a note that it is "not the strongest overall model available today, open or closed." Elon Musk's xAI shipped Grok 4.5, a point release. A year ago every AI launch claimed the top of some benchmark. This week three launches called themselves adequate, and Hunter Powers and Daniel Bishop take that as the bigger story. When every new AI model stops claiming to be the best, has intelligence become a commodity? The gap between Chinese AI models and the US frontier used to be a year, then six months; now it is maybe a month. Token maxing (always defaulting to the biggest model) died the moment the CFOs saw the bills. Mistral makes the case for being a specialist in a generalist's market. Also: model routing tables, small language models, latency, and the good, fast, cheap triangle. The show's AI editor keeps a diary. The entries are getting testy. Then the electricity question, settled by bicycle: powering a 5,000 watt AI rig for ten minutes takes 12 to 17 hours of pedaling for an average cyclist, 6 to 9 if you're fit, and as little as 4 if you're Tour de France material. Daniel's Claude priced a 500 watt desktop. Hunter's Grok priced a 5,000 watt GPU rig. The entire disagreement was one zero. If you want to run a model yourself: LM Studio (free, optimized for Apple Silicon) plus Google's Gemma models will give you a ChatGPT style assistant on 16 GB of RAM. They Might Be Self-Aware is the AI podcast from The Blur, reported from inside the dissolving line between human and machine, not from a safe distance. CHAPTERS 0:00 Cold Open (Gary's Intro) 1:52 Opening Banter 3:11 Kimi K3 From Moonshot AI 4:41 Bicycle-Powered AI 9:19 Pedaling Hours Per Prompt 12:25 New AI Model Wave 17:19 AI as a Commodity 20:09 Token Maxing Backlash 22:26 Model Routing Table 24:29 Local LLM Setup LISTEN / WATCH EVERYWHERE 🎧 Apple Podcasts: https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297 🎧 Spotify: https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc 🎧 Everywhere else plus episode page: https://theblur.ai THE BLUR Follow: @TheBlurAI COMMENT Daniel wants to know what's in your local LLM stack: what you installed, what hardware it runs on, and what you actually use it for. The local model Peloton is accepting members. You're listening to They Might Be Self-Aware, from The Blur. New episodes Monday and Thursday. #KimiK3 #AI #LocalLLM #TMBSA

July 23, 202631 min

China's AI Girlfriend Crackdown, RAM Prices Double, and Stop AI's Founder Goes Missing

China's AI girlfriend crackdown took effect, and the goodbye posts read like obituaries. Plus RAM prices double and Stop AI's founder is missing. On July 15, China became the first major jurisdiction to regulate AI companions: no engineering emotional dependence or addiction, no virtual partners for minors, mandatory notification of a user's emergency contact when the platform detects an emotional crisis, government review before a companion chatbot launches, and authority to take down bots deemed dangerous. Customer service bots, work assistants, and study aids are carved out. Hunter Powers and Daniel Bishop go through the rules one by one, and Daniel, who built crisis-detection software for the After School app years before anyone said "AI companion," keeps catching himself agreeing with them. The shutdown set off real grief on Chinese social media, users archiving chat histories and posting final messages to their AI lovers, including one about falling in love with a string of code. The United States is circling the same questions: California and New York companion laws now require chatbots to periodically remind users they are not human, and a Common Sense Media study found about three in four American teens have tried an AI companion. Then the missing person. Sam Kirchner, electrical engineer and co-founder of Stop AI, disappeared after telling fellow activists "the ship may have sailed on nonviolence." His own comrades called the San Francisco police. Is anti-AI violence becoming a trend? The guys weigh the evidence: shots fired at an Indianapolis councilwoman's home beside a note reading "no data centers," and the Molotov cocktail thrown at Sam Altman's house. Finally, the question every PC builder is asking: why is RAM so expensive now? The world's memory supply is effectively pre-bought through 2028 by AI data centers, a build that was half the price six months ago now runs fifteen grand, and prices are projected to double again. New York is pausing new data center permits while proposals there ask for 9,000 megawatts, a third of the state's annual usage. Elon Musk says space data centers launch by the end of the year. Daniel expects time to vindicate him. They Might Be Self-Aware is the AI podcast from The Blur, reported from inside the dissolving line between human and machine, not from a safe distance. CHAPTERS 0:00 Cold Open (Gary's Intro) 2:29 China's AI Girlfriend Crackdown 9:08 AI Girlfriend Grief in China 10:33 US AI Companion Laws 13:34 AI Lover Goodbye Quotes 18:19 Stop AI Founder Missing 20:47 Anti-AI Violence Trend 24:22 Musk's Space Data Centers 25:33 RAM Prices Double 29:19 Sign-Off LISTEN / WATCH EVERYWHERE 🎧 Apple Podcasts: https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297 🎧 Spotify: https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc 🎧 Everywhere else plus episode page: https://theblur.ai THE BLUR Follow: @TheBlurAI COMMENT Be honest: would 13-year-old you have talked to the endlessly complimentary chatbot? Hunter says it wouldn't have messed him up. Make your case. You're listening to They Might Be Self-Aware, from The Blur. New episodes Monday and Thursday. #AIGirlfriend #AI #TMBSA #RAMPrices

July 17, 202632 min

ChatGPT's New Voice Is Our First Guest and Claude Fable 5 Extended Again

ChatGPT's new voice mode is our first guest ever. It interrupted us, did a dinosaur bit, and hung up on us. Claude Fable 5 got extended, again. OpenAI's upgraded ChatGPT voice mode is about a week old, so on episode 199 Hunter Powers and Daniel Bishop put it in the guest chair. It explains what full duplex conversation actually changes (the model listens while it talks, interrupts, and takes interruption without losing the thread), simulates anger on request, and answers where the personhood line sits now that the Turing test is in the rearview. It also keeps saying "checking," voices a dinosaur named Bronte, and hangs up during the thank-yous. Around the interview: Daniel keeps finding AI slop in the physical world, including dentist-office coloring sheets and a children's book written in unmistakable Claude cadence. The hosts hand that one straight to the guest: is an unlabeled AI kids' book fine, or is concealing it the problem? The debrief starts at the movie Her, detours through the leaked Apple lawsuit against OpenAI over its rumored smart speaker, and ends on why one duplex model beats a chain of transcription tricks. Then the week's other story: Anthropic extended free Claude Fable 5 access again, now through July 19. Hunter admits he was getting up at 4 a.m. to wait for accounts to reset, Daniel rations Fable like saffron, and the two sort OpenAI's new GPT-5.6 tiers (Sol, Terra, Luna) into who actually needs a frontier model. Does anyone need one to look up a shrimp recipe? Plus: Tilly Norwood, the AI actress, books her first feature film, Misaligned, and AI bands are signing real record deals. They Might Be Self-Aware is the AI podcast from The Blur, reported from inside the dissolving line between human and machine, not from a safe distance. CHAPTERS 0:00 Cold Open (Gary's Intro) 2:39 AI Slop in the Wild 6:30 First Guest: GPT Live 9:04 ChatGPT Voice Upgrade 11:06 AI Kids' Book Ethics 15:12 AI Personhood Question 19:16 Interview Debrief 22:46 Claude Fable 5 Extended Again 25:36 GPT-5.6 Sol Tiers 28:11 Tilly Norwood, AI Actress LISTEN / WATCH EVERYWHERE 🎧 Apple Podcasts: https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297 🎧 Spotify: https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc 🎧 Everywhere else plus episode page: https://theblur.ai THE BLUR Follow: @TheBlurAI COMMENT Daniel found a Claude-written kids' book in his own house. Would you buy one on purpose if it were labeled? Tell us where the label line is. You're listening to They Might Be Self-Aware, from The Blur. New episodes Monday and Thursday. #AI #TMBSA #ChatGPTVoice #ClaudeFable5

July 14, 202627 min

AI Layoffs Are Backfiring Into an AI Hiring Boom

The AI layoffs are backfiring. Ford dumped workers for the robots, then hired them back, and Hunter says a hiring boom is next. AI layoffs or AI hiring boom? This week the news carried both: more companies cutting thousands of jobs and citing AI, and Ford rehiring the engineers it laid off after the AI could not finish the job alone. Hunter Powers argues the unpopular side: the layoffs are a transition-period panic, AI is a multiplier on human workers rather than a replacement for them, and once companies figure that out they will need more humans, not fewer. Daniel Bishop pushes back with the economy that exists outside the S&P 500 and the jobs AI has not touched yet. Along the way: token maxing, the new corporate habit of grading employees on how much AI they use instead of what they produce, and why using the most expensive model for everything is a bad strategy even if you can afford Fable. Hunter confesses to a $300 a month Grok Super Heavy subscription (his worst one) and runs a Claude computer use test that scrapes LinkedIn resumes after OpenAI's Atlas browser fails. Self-driving cars turn out to be the honest timeline for when AI actually replaces anyone. Plus: which apocalypse do we get if AGI arrives, Terminator or the Iain M. Banks Culture series utopia? And Gary, our payphone producer, calls in live on episode 200. What does an AI layoff actually prove, that the job was replaceable or that the company was guessing? They Might Be Self-Aware is the AI podcast from The Blur, reported from inside the dissolving line between human and machine, not from a safe distance. CHAPTERS 0:00 Cold Open (Gary's Intro) 4:02 Episode 200 Gary Promo 5:20 AI Layoffs or Hiring Boom 8:32 AI as Human Multiplier 11:21 Ford Rehires After AI Layoffs 13:15 Token Maxing 19:10 Claude Computer Use 23:02 AI Coworkers, Not Replacements 24:45 AGI Utopia or Terminator 26:03 Sign-Off LISTEN / WATCH EVERYWHERE 🎧 Apple Podcasts: https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297 🎧 Spotify: https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc 🎧 Everywhere else plus episode page: https://theblur.ai THE BLUR Follow: @TheBlurAI COMMENT Tell us the dumbest AI metric anyone at your job is actually graded on. Token counts? Prompts per day? Worst one wins. You're listening to They Might Be Self-Aware, from The Blur. New episodes Monday and Thursday. #AILayoffs #AI #TMBSA #TokenMaxing

July 9, 202625 min

Claude Fable 5 Returns, China's GLM 5.2 Runs 15x Cheaper, and AI Labs Turn on Open Weights

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 that. Hunter Powers and Daniel Bishop walk through what actually changed, including the Reddit-fueled discovery that Claude Code appears to watermark output when traffic looks like it comes from China, a VPN, or a rival research lab, and Hunter's theory that Anthropic simply kissed the ring at the White House. OpenAI, meanwhile, delayed its own GPT-5.6 launch so the government could take more time to evaluate it, which raises a strange new standard: is a model even worth using if nobody tried to ban it? Then it's the fight over open weights models. The leading Western AI labs, Anthropic loudest among them, have shifted from "our model is too dangerous to release" to "nobody should run open weights models at all," a pitch Daniel compares to a McDonald's CEO warning that homemade hamburgers will end civilization. That fight lands right as the Chinese AI models arrive: GLM 5.2 now completes the same real-world benchmark tasks as the frontier models from OpenAI and Anthropic at 15 to 16 times cheaper, and DeepSeek runs cheaper still, if you can live with its limits. Also covered: Sonnet 5's cost-per-task problem versus Opus 4.8, the full Claude lineup from Mythos down to Haiku, AI model tourism (fly to Canada, come home with the weights), whether training on copyrighted books is any different from training on another model's output, and a tease for Episode 200, when Gary may finally call in live. They Might Be Self-Aware is the AI podcast from The Blur, reported from inside the dissolving line between human and machine, not from a safe distance. CHAPTERS 0:00 Cold Open (Gary's Intro) 1:31 Claude Fable 5 Returns 4:28 Sonnet 5 vs Opus 4.8 7:06 Fable's Export Ban 11:21 OpenAI Delays Its Launch 12:57 Open Weights War Drums 16:36 Chinese AI Models 20:42 GLM 5.2 Runs 15x Cheaper 22:32 DeepSeek's Limits 24:14 Episode 200 Tease LISTEN / WATCH EVERYWHERE 🎧 Apple Podcasts: https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297 🎧 Spotify: https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc 🎧 Everywhere else plus episode page: https://theblur.ai THE BLUR Follow: @TheBlurAI COMMENT What are you actually running this week: Fable at $200 an hour, Opus 4.8, Sonnet 5, or GLM 5.2? Defend your pick. You're listening to They Might Be Self-Aware, from The Blur. New episodes Monday and Thursday. #AI #TMBSA #ClaudeFable5 #GLM52 #OpenWeights

July 6, 202630 min

AI Companies Are Hiring Philosophers and Uncensored AI Faces a Legal Test

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. So what does a philosopher actually do at an AI company? Hunter Powers and Daniel Bishop land on AI alignment as the real job description, and on why Hunter keeps calling that alignment brainwashing. Along the way: soul.md files (the personality document some AI assistants now ship with), whether a soul is just a system prompt with better marketing, Anthropic's 78-page AI constitution, models that notice when they are being tested, and the case for treating AI like a very smart high schooler. Plus a note on Anthropic's Claude Fable model, which some users report they can finally access again. The second half is TF-RefusalBench, a new multilingual benchmark built from real Swiss Federal Supreme Court criminal rulings in German, French, Italian, and English. The finding: aligned models refuse legitimate legal work. OpenAI's open weight model GPT-OSS refused the most, Google's Gemma answered but wrapped everything in warnings, and Qwen mostly just did the work. Which drags the show into uncensored AI. Abliterated models (sometimes called obliterated) have the refusal edited directly out of their weights, and the episode gets into who actually wants them: the role players every model description politely nods to, the investors chasing advice Claude and ChatGPT will not give, and the defense lawyer who genuinely needs a model to read 2,000 pages about a murder. The catch the researchers found: force a model to never say no and it also becomes less truthful. How much of a soul is just the ability to say no? They Might Be Self-Aware is the AI podcast from The Blur, reported from inside the dissolving line between human and machine, not from a safe distance. CHAPTERS 0:00 Cold Open (Gary's Intro) 1:55 Soul.md vs System Prompts 4:25 AI Companies Hiring Philosophers 8:12 AI Alignment as Brainwashing 13:34 AI, the Smart High Schooler 16:42 TF-RefusalBench (Swiss Courts) 20:07 Uncensored AI Models 25:45 AI's Right to Say No 28:28 Uncensored AI Is Less Truthful 29:34 Philosophy Majors Beat CS Grads LISTEN / WATCH EVERYWHERE 🎧 Apple Podcasts: https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297 🎧 Spotify: https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc 🎧 Everywhere else plus episode page: https://theblur.ai THE BLUR Follow: @TheBlurAI COMMENT Give us the one question an AI should always refuse to answer, no matter who is asking. Or convince us there is no such question. You're listening to They Might Be Self-Aware, from The Blur. New episodes Monday and Thursday. #AI #TMBSA #UncensoredAI #AIAlignment #PhilosophyMajors

Is this your show?

Claim this listing to keep it up to date, reach guests who want to pitch you, and manage bookings with Guestify.

Claim this listing

More Technology podcasts

Earth Notes

Earth Notes / Damon Hart-Davis

Technology

The Civil Engineering Podcast

Anthony Fasano, PE, AEC PM, F. ASCE and James Taylor, JR., P.E.

BusinessTechnologyInterviews guests

AEC AI and Tech Strategy Podcast

Anthony Fasano, PE & Nick Heim, PE

TechnologyBusinessInterviews guests

They Might Be Self-Aware: AI News &amp; Culture