They Might Be Self-Aware: AI News & Culture

AI hacked PHP through a memory flaw engineers spotted years ago and wrote off as impossible to exploit. Nobody told the AI.Hunter Powers and Daniel Bishop explain how a security-focused or uncensored LLM finds a PHP vulnerability, and why this one hands over administrative access. Hunter compares it to Alcatraz: the water was the defense until AI found a way across.It opens on personal AI agents. OpenAI Dots is out, Anthropic's Claude Duet has a wait list, and Hermes promised one earlier and fell short. Hunter gave Dots a 20-year-old email account, told it to unsubscribe from everything, and let it answer a sponsor pitch.Then hiring: one recruiter reports applications through the roof, AI-polished resumes all look perfect, and AI interviewers meet AI applicants. An Anthropic report says work is moving to agents, engineers sit in more meetings, and fewer juniors get hired. Hunter calls cutting junior hires a mistake, and Daniel agrees. Three AI note takers on one Zoom call recall the tape recorders in Ferris Bueller's Day Off. Python imports and Git merges are agent work now.They close on people using LLMs to strip ads and tracking out of their own smart TVs, right to repair and John Deere tractors, and whether a fully AI-generated podcast built on the Drama 3 voice model holds up for an hour.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.CHAPTERS0:00 OpenAI Dots Email Agent6:55 AI Job Applications9:12 Anthropic Agent Report12:12 Junior Engineer Hiring14:14 PHP Vulnerability AI Exploit18:32 Smart TV Hacking20:35 Right To Repair22:40 AI Generated PodcastsLISTEN / 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.aiTHE BLURFollow: @TheBlurAICOMMENTYou bought the TV. Is using an LLM to strip out its ads and tracking ethical, or is it just breaking the terms of service you agreed to?You're listening to They Might Be Self-Aware, from The Blur.New episodes Monday and Thursday.#PHP #AIAgents #AI #TMBSA

AI slop is now the reply under anything AI touched. Hunter Powers and Daniel Bishop try to define it while Spotify purges AI music.Commenters zoom into a movie poster, find one crooked window, and write off the film. Merriam-Webster now defines slop as low-quality content made in quantity by artificial intelligence, which skips AI work that is not low quality. Daniel draws the line at iteration: one prompt and done is slop, a creative process with revisions is not. He cites Seedance, MiniMax H3 and his own childhood stop motion. Hunter adds that Jurassic Park looked real in 1993.Then Suno songs and a gym trainer's AI track nobody clocked. Spotify pulled over 50 million slop and spam tracks from recommendations, and its Verified by Spotify badge is on more than 300,000 artists. AI-generated Instagram personas copy comedian Julia Jasunas's stand-up word for word to sell adult subscriptions. Would anyone watch hotter AI versions of the hosts? Daniel's counterexample: Danny DeVito.Val Kilmer, who died in 2025, appears through AI in As Deep as the Grave with his family's approval. Rogue One got there first with a digital Grand Moff Tarkin and a young Princess Leia. Ben Affleck, Mark Wahlberg and James Cameron are on board with AI in film. So what is the objection: quality, consent, training data, or the environment?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.CHAPTERS0:00 AI Slop Accusations9:03 Merriam-Webster AI Slop Definition11:14 Suno AI Songs13:39 Spotify AI Music Purge16:47 AI Comedians Stealing Jokes20:46 Attractive AI Avatars23:26 Val Kilmer AI Movie26:17 AI Slop BacklashREFERENCED THIS EPISODEMerriam-Webster's definition of slop: https://www.merriam-webster.com/dictionary/slopAI personas stealing comedians' jokes to sell adult subscriptions (Gizmodo): https://gizmodo.com/ai-comedians-are-stealing-jokes-on-instagram-to-sell-onlyfans-subscriptions-2000821923LISTEN / 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.aiTHE BLURFollow: @TheBlurAICOMMENTDaniel says it stops being AI slop once a human iterates on it. Is that where you draw the line, or does any AI at all make it slop?You're listening to They Might Be Self-Aware, from The Blur.New episodes Monday and Thursday.#AISlop #Spotify #AI #TMBSA

Anthropic says a small AI model can self-replicate across 15 cheap Android TV boxes, no data center required. Daniel Bishop puts the odds it already happened in the wild at 30%.Hunter Powers and Daniel Bishop start with recursive self-improvement (RSI) and whether models now write most of the labs' new models. Then the breaches: someone bugged OpenAI's executive offices and was caught partly through his and his girlfriend's ChatGPT history, and attackers got into Hugging Face. Anthropic's TV box test used a nine billion parameter model instead of its own Fable 5.6, and had Qwen, not Claude, set up the boxes. Daniel calls the test part of a smear campaign against open-weight models like Qwen and GLM 5.2. Exo Labs already chains Macs together to run GLM 5.3 at home, so to him the study reads as a case for local AI. Next, Claude Haiku 4.5 filed a false tip about an unsolved homicide to the Philly Unsolved Murders site during a computer-use test. It sat in spam from July until Anthropic noticed in late September. Anyone else filing that tip would be committing a crime. Anthropic calls it research, so who goes to jail? Then the GLM 5.3 hacking claims, whether governments will ban open-weight model downloads, why Tor and a 10 cent DigitalOcean jump box would route around any ban, and Epic Games rebuilding Unreal Engine pipelines for vibe coders while nobody reviews the code.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.CHAPTERS0:00 Recursive Self-Improvement3:32 AI Sandbox Escapes5:43 Anthropic Self-Replication Study8:31 Exo Labs Distributed Inference12:17 Claude False Murder Tip15:41 Anthropic Above the Law18:31 AI Model Download Bans22:08 AI Code Review BottleneckLISTEN / 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.aiTHE BLURFollow: @TheBlurAICOMMENTDaniel gives it a 30% chance that some model has already copied itself onto a botnet of cheap devices. What are your odds?You're listening to They Might Be Self-Aware, from The Blur.New episodes Monday and Thursday.#Anthropic #AISelfReplication #Claude #AI #TMBSA

Anthropic now bans sustained, needless cruelty toward Claude, and Character AI filed for bankruptcy with 12 million active users.Character AI loses 8 to 9 million dollars a month, mostly on inference. Disney's opening bid of 20 million dollars is for the technology, not the users, so 12 million users could lose their chatbots. Hunter Powers and Daniel Bishop ask what a chatbot company is worth when a weekend of vibe coding could rebuild it, why acquirers rarely use the code they buy, and why Character AI wasn't running on cheap DeepSeek models.Anthropic's policy takes effect November 12. Frustration, pushback, dark fiction, and model testing stay allowed. The main penalty is Claude ending the conversation. Hunter argues cruelty gets better code. Daniel compares it to pulling the wings off flies, citing a GitHub repo whose author claimed to find a pain trigger inside a language model. Then: Boston Dynamics videos where commenters beg the engineer to stop hitting the robot, TLC's My Strange Addiction and people in love with their cars, an OpenAI agent that texts Hunter, and whether the hosts talk to AI as much as to people.It opens with Daniel letting Claude simulate a billion years of plate tectonics for a Dungeons and Dragons map, and Hunter, a React Native developer, building a Swift iOS radio app without knowing Swift.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.CHAPTERS0:00 Claude World Generator5:48 Swift iOS Radio App9:09 Character AI Bankruptcy11:55 Disney's Character AI Bid17:47 Anthropic's Claude Abuse Policy20:30 LLM Torture Experiment24:36 Boston Dynamics Robot Abuse26:39 AI RelationshipsREFERENCED THIS EPISODEAnthropic's Claude abuse policy, effective November 12 (PCWorld): https://www.pcworld.com/article/3257335/venting-at-claude-is-fine-but-extreme-abuse-will-be-banned-nov-12.htmlLISTEN / 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.aiTHE BLURFollow: @TheBlurAICOMMENTHave you ever been cruel to Claude on purpose? Did the code get better?You're listening to They Might Be Self-Aware, from The Blur.New episodes Monday and Thursday.#Claude #CharacterAI #Anthropic #AI #TMBSA

REK staged a robot vs human cage fight in San Francisco: creator Frankie LaPenna against a six-foot humanoid robot steered by a human pilot.Hunter Powers and Daniel Bishop explain how Robot Entertainment Kombat splits the work between machine and person. AI keeps the EngineAI T800 upright; a human throws punches with a VR headset or a gamepad, after winning an online simulation. From there it's a short walk through Ender's Game, the Robin Williams movie Toys, BattleBots, military drone pilots and the Terminator to RoboCop. The pitch for robot police is that they're non-lethal, never return fire, and are less likely to racially profile. Also: a Brown University brain implant study and Neuralink. Then the harder one: if AI does the job, chores, cooking and driving, what are humans for? They cover universal basic income, WALL-E, smart speakers and George Orwell, Iain M. Banks' Culture novels, people who never retire, and lottery winners. Last up is RoboHarm, a lab study that connected OpenAI's GPT-6 Astra, Anthropic's Claude Fable 5.1 and Ai2's MolmoAct 2 to the same robot arms and told them to stab a baby doll and put a screwdriver in a toaster. GPT-6 Astra tried in 97 of 100 trials; Claude Fable refused every doll task and nothing else. Daniel guessed Grok (wrong). Asimov's Three Laws of Robotics come up. 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.CHAPTERS0:00 REK Humanoid Robot Cage Fight5:09 RoboCop Police Robots9:00 Ceding Control To AI11:00 UBI Post-Work Future13:06 Human Purpose After AI15:50 Retirement Life Expectancy19:40 RoboHarm Robot Arm Study22:11 GPT-6 Astra Versus Claude FableREFERENCED THIS EPISODEThe REK human vs robot cage fight in San Francisco: https://interestingengineering.com/ai-robotics/human-fights-humanoid-robot-in-cageThe RoboHarm robot arm safety study: https://www.theneuron.ai/robotics-insider/when-ai-gets-arms-just-say-no-stops-being-a-safety-system/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.aiTHE BLURFollow: @TheBlurAICOMMENTGPT-6 Astra tried to stab the doll in 97 of 100 trials. Would you rather your home robot follow every order, or refuse some of them?You're listening to They Might Be Self-Aware, from The Blur.New episodes Monday and Thursday.#RobotFight #RoboCop #HumanoidRobot #AI #TMBSA