They Might Be Self-Aware

Did Anthropic Just Declaw Claude Fable 5?

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 hacking, biology and chemistry, and anything that helps you train a competing model. Ask Fable 5 a security question and it does not refuse, it quietly hands the request down to Claude Opus, a weaker model, and lets that one say no. We get into why a lab would defang its own best model right before going public, what that says about protecting the moat, and whether Fable 5 is worth roughly twice the price of Opus 4.8 when heavy use burns two hundred dollars an hour. Then the part that hits your job. Both hosts handed Fable 5 a goal, walked away, and came back to finished software. Hunter built a playable liminal-space data center game with a live DeepSeek integration. Daniel built a Sisyphus mobile game, art and all, wired to Google's Nano Banana Pro. Anthropic says Fable 5 autonomously migrated fifty million lines of code for Stripe. So is software engineering dead? Daniel calls it, period, full stop, and we argue about what that leaves for the people who used to write the code. We also get into the one-million-token context window and why you compact it early, Cursor's benchmark putting Fable 5 first in quality and first in cost, and AI budgets as a line item in your next salary negotiation. Plus the case for using a frontier model to orchestrate cheaper Opus, Sonnet, and DeepSeek sub-agents, and whether self-awareness comes from one giant model or a committee of small ones. 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 Anthropic Ships Fable 3:00 Fable's Filed-Down Teeth 4:46 Anthropic's Moat 11:46 Fable's Real Cost 15:40 Fable's Autonomy Leap 21:39 Building Games With Fable 28:41 Software Engineering Is Dead 29:41 AI Budget as Compensation 34:58 Self-Aware by Committee REFERENCED THIS EPISODE Anthropic's Claude Fable 5 and Mythos 5 announcement: https://www.anthropic.com/news/claude-fable-5-mythos-5 [https://www.anthropic.com/news/claude-fable-5-mythos-5] LISTEN / WATCH EVERYWHERE 🎧 Apple Podcasts: https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297 [https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297] 🎧 Spotify: https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc [https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc] 🎧 Everywhere else plus episode page: https://theblur.ai [https://theblur.ai] THE BLUR Follow: @TheBlurAI COMMENT Daniel says traditional software engineering is dead, period, full stop. If Fable 5 can ship the whole app overnight, what is the one thing you still would not let it touch? Be specific. You're listening to They Might Be Self-Aware, from The Blur. New episodes Monday and Thursday. #ClaudeFable5 #Anthropic #SoftwareEngineering #AI #TMBSA

Anthropic Beat OpenAI to the IPO

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 companies air their own dirty laundry, the way Uber once admitted it might never have a profitable business model), why companies file confidentially, and why whoever publishes their numbers first takes the hardest media hit. Then the bigger question: is this the AI bubble's cash-out moment, the five people at the top each collecting their seventy-two million before it comes down, or a sign the industry is maturing? They get into investor pressure after Anthropic's $65 billion raise, how startup dilution actually works, the expert consensus that all three IPOs pop short term and trade down within a year, and where the pension funds and 401k money fit in. From there: whether OpenAI and Anthropic have any real moat, ChatGPT's claimed one billion monthly users (per Sensor Tower) versus Claude's 56 million, the everything-company problem, Google's free local Gemma 4 models, Qwen 3.6, and DeepSeek V4 at one-hundredth the cost. If a free model on a normal laptop does most of what the $200-a-month plan does, what exactly is a trillion-dollar valuation buying? Plus: enterprise AI spend failing the ROI math, the quiet walkback of "AI will take 80% of engineering jobs," what AGI would do to all of these valuations, and why an AI employee (ask anyone running Nous Research's Hermes agent) still costs more than the human it replaces. 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) 3:17 Anthropic's IPO Filing 7:32 SpaceX IPO 10:01 AI Bubble Cash-Out 15:44 IPO Pop Predictions 18:00 OpenAI's Missing Moat 23:48 Google's Gemma 4 27:02 OpenAI vs Free Models 31:46 Enterprise AI ROI 36:19 The AGI Question 39:11 AI Employees Cost More LISTEN / WATCH EVERYWHERE 🎧 Apple Podcasts: https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297 [https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297] 🎧 Spotify: https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc [https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc] 🎧 Everywhere else plus episode page: https://theblur.ai [https://theblur.ai] THE BLUR Follow: @TheBlurAI COMMENT Check your 401k. Are you glad to see SpaceX and Anthropic show up in there, or are you hunting for the "not AI" fund? Tell us which one you are. You're listening to They Might Be Self-Aware, from The Blur. New episodes Monday and Thursday. #AI #TMBSA #Anthropic #OpenAI #SpaceX

AI Is Already Inside California's Courtrooms

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 like in 2026: a tool that reads thousands of similar cases, flags sentencing outliers, and surfaces Racial Justice Act petitions where a defendant got a harsher term than the data supports. They start from the Freakonomics hungry-judge study (rulings swing softer right after lunch), which raises the real question: if an AI sentencing assistant is measurably more consistent than a tired human judge, do we owe defendants the machine? Then it gets messier. Lawyers are already being disbarred for filing ChatGPT briefs with fake citations. A defense AI against a prosecution AI turns the courtroom into a GPU arms race. And the same logic that smooths out a biased sentence can quietly delete the human discretion we only miss once it is gone. The second half follows the surveillance thread the courtroom opens. AI surveillance is moving from passive recording to natural-language search: Flock cameras shared across police departments and private owners, China's ChatGPT-style interface for querying a whole city's camera network, pre-crime prediction, and WiFi sensing that reconstructs people through walls (yes, it is on GitHub). Plus Gaussian splatting that rebuilds a room, or a person, from a couple of photos. Where do we let AI in, and where do we draw the line? 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) 2:57 Chef Claude 4:46 California Courts: Learned Hand 5:42 Algorithmic Bias 11:04 The Hungry Judge 12:43 Robot Wardens 14:19 Smoothing the Outliers 18:34 AI Surveillance 22:10 Flock Cameras 27:47 WiFi Through Walls LISTEN / WATCH EVERYWHERE 🎧 Apple Podcasts: https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297 [https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297] 🎧 Spotify: https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc [https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc] 🎧 Everywhere else plus episode page: https://theblur.ai [https://theblur.ai] THE BLUR Follow: @TheBlurAI COMMENT If an AI sentencing tool is provably more consistent than a hungry, tired, biased judge, would you want it deciding your sentence, or is the flawed human the whole point? You're listening to They Might Be Self-Aware, from The Blur. New episodes Monday and Thursday. #AIinCourt #AI #TMBSA #AISurveillance

The Pope's 40,000-Word AI Verdict: Not Human.

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 finding internal states in AI models that functionally mirror joy, satisfaction, fear, grief, and unease. Hunter Powers and Daniel Bishop set the two claims side by side: if a machine acts scared, does it matter whether the fear is real? Then the Center for AI Safety's Wellbeing Index: 56 AI models, about 500 conversations each, ranked by functional wellbeing. Grok tested happiest. A Gemini model came in last. Hunter floats a theory about why the least aligned model might be the most cheerful, and Daniel stress-tests it. The back half is education. AI homework cheating has made the essay, the take-home test, and the admissions letter trivially gameable. Mount St. Vincent religious studies professor Jane Sloan Peters got choked up in front of her class: students no longer struggle through her Letters from Prison course, and she grieves it. School has spent thousands of years grading outputs, and AI just solved outputs. Daniel gives the current system ten years. Hunter proposes grading the struggle instead. Idiocracy comes up, affectionately. 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) 2:36 The Pope's 40,000-Word Encyclical 4:53 Pope's Verdict: Not Human 8:23 Anthropic's AI Emotions 13:10 Anthropic's Religious Outreach 16:58 AI Wellbeing Index 18:06 Grok, the Happiest AI 22:36 AI Homework Cheating 27:53 Gaming the Education System 33:44 Education After AGI REFERENCED THIS EPISODE Pope Leo XIV's AI encyclical, the New York Times report: https://www.nytimes.com/2026/05/25/world/europe/pope-leo-encyclical.html [https://www.nytimes.com/2026/05/25/world/europe/pope-leo-encyclical.html] Chris Olah's remarks on the encyclical (Anthropic): https://www.anthropic.com/news/chris-olah-pope-leo-encyclical [https://www.anthropic.com/news/chris-olah-pope-leo-encyclical] The Center for AI Safety's AI Wellbeing Index (Fortune): https://fortune.com/2026/05/07/researchers-ai-models-drugs-euphoric-dysphoric/ [https://fortune.com/2026/05/07/researchers-ai-models-drugs-euphoric-dysphoric/] Jane Sloan Peters on grieving what AI took from learning (Daily Nous): https://dailynous.com/2026/05/01/grieving-what-ai-has-taken-from-learning/ [https://dailynous.com/2026/05/01/grieving-what-ai-has-taken-from-learning/] LISTEN / WATCH EVERYWHERE 🎧 Apple Podcasts: https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297 [https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297] 🎧 Spotify: https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc [https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc] 🎧 Everywhere else plus episode page: https://theblur.ai [https://theblur.ai] THE BLUR Follow: @TheBlurAI You're listening to They Might Be Self-Aware, from The Blur. New episodes Monday and Thursday. #AI #TMBSA #PopeLeoXIV #Anthropic #Grok

Claude Just Cracked 11 Years of Lost Bitcoin

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 notebook and computer files, and the model narrowed an effectively infinite seed-phrase search down to a brute-force range small enough to win. No protocol was broken. Does it matter that Claude did not break Bitcoin if the wallet opened anyway? Hunter has his own version of the nightmare, the roughly eight Bitcoin he mined at $12 a coin and stranded on an old SATA drive, gone. From there the bigger question: how much of your life do you hand an AI? OpenAI is now wiring ChatGPT into Plaid, the same pipe that connects Schwab, Fidelity, Chase, and 12,000 other institutions, so the model can read your bank account and flag your dead subscriptions. Daniel's working definition of AGI is the day it can file your taxes start to finish. Then the money story: Sam Altman is offering Y Combinator startups $2 million in OpenAI credits in exchange for equity, and we argue over whether that is ordinary venture capital or OpenAI buying a look at how the next wave of startups actually builds with AI. Plus Google's Gemini 3.5 Flash, the model the benchmarks loved and reviewers did not, and the benchmaxing accusation trailing it; Google's new Spark assistant; and Daniel's case that the future is local AI running on hardware in your house, if you can still afford a graphics card. 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) 1:21 Opening Banter 2:26 Hunter's Lost Bitcoin 5:38 Claude Bitcoin Recovery 9:24 Teacher Email Hack 11:59 AI Bank Access 17:16 OpenAI's $2M Startup Deal 25:15 Gemini 3.5 Flash 29:27 Local AI Future REFERENCED THIS EPISODE The Claude Bitcoin wallet recovery story (crypto.news): https://crypto.news/claude-helps-man-recover-5-bitcoin-after-old-wallet-search/ [https://crypto.news/claude-helps-man-recover-5-bitcoin-after-old-wallet-search/] LISTEN / WATCH EVERYWHERE 🎧 Apple Podcasts: https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297 [https://podcasts.apple.com/us/podcast/they-might-be-self-aware/id1730993297] 🎧 Spotify: https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc [https://open.spotify.com/show/3EcvzkWDRFwnmIXoh7S4Mb?si=3d0f8920382649cc] 🎧 Everywhere else plus episode page: https://theblur.ai [https://theblur.ai] THE BLUR Follow: @TheBlurAI COMMENT Hunter would let an AI read his bank account but never touch his crypto wallet. Where exactly is your line, and what is on the wrong side of it? You're listening to They Might Be Self-Aware, from The Blur. New episodes Monday and Thursday. #Claude #Bitcoin #LostBitcoin #AI #TMBSA