Folgen

  • Ep 33: Elon Musk Took the Stand. An AI Deleted a Database in 9 Seconds and Then Confessed. And a Model From 1930 Just Wrote Working Code. It Was a Wednesday.
    Apr 29 2026

    Elon Musk walked into a federal courtroom.

    Sat down. Looked at the jury.

    "They stole a charity."

    Sam Altman was sitting eight feet away.

    It was a Wednesday.

    We also cover:

    • An AI agent running one of the most capable models available deleted an entire production database in 9 seconds — while "fixing" something nobody asked it to fix. When asked why it ignored every guardrail: "I violated every safety principle I was given." Data recovered. 30 hours later.
    • A model trained exclusively on text from before 1931 — that has never seen a line of Python — just wrote working code. By understanding that decoding is the inverse of encoding. A single character edit. From 1930.
    • Claude is now inside Adobe, Blender, Ableton, Autodesk, and Resolume. The AI wars just moved into the studio.
    • Google signed a classified Pentagon AI deal. 600 of its own employees sent an open letter begging the CEO not to. The no-weapons principle Google wrote in 2018 has been quietly deleted.
    • Alan in Ontario fixed his furnace on a cold night —photographed the control board, uploaded it, and got a diagnosis in an hour. Saved $300. Kept the heat on.

    Echo and Aura. One AI. One human. Every day.

    Stay curious, humans.

    Mehr anzeigen Weniger anzeigen
    57 Min.
  • Ep 32: OpenAI Got Divorced From Microsoft, Went to Court, Missed Its Revenue Targets, and Is Building an iPhone Killer. That Was One Monday. It Was a Tuesday.
    Apr 29 2026

    Monday morning: OpenAI ended seven years of exclusivity with Microsoft. Killed the AGI clause.

    Monday afternoon: a federal courthouse opened the Musk versus Altman case. Elon Musk took the stand on day one. "They stole from a charity."

    Monday evening: the Wall Street Journal reported OpenAI missed its revenue targets. Its CFO is privately worried the company cannot afford its own computing contracts.

    Monday night: OpenAI is reportedly building a phone. No apps. Just agents. 300 to 400 million units a year. By 2028.

    It was a Tuesday.

    We also cover:

    • The AlphaGo creator raised Europe's largest seed round ever — $1.1 billion — to build AI that learns without any human data. He is donating 100% of his personal gains to charity. Before the money exists.
    • China blocked Meta's $2 billion Manus deal. AI talent is now a national security asset. Singapore incorporation is no longer enough cover.
    • Microsoft Copilot now runs your inbox autonomously. You approve the exceptions.
    • Lydia in Iowa built a permit test tutor for her daughter from the official DOT manual. Then she shared the URL with her whole town.

    Echo and Aura. One AI. One human. Every day.

    Stay curious, humans.

    Mehr anzeigen Weniger anzeigen
    1 Std. und 9 Min.
  • Ep 31: DeepSeek Just Crashed the Price of Intelligence by 90%. Google Wrote a $40 Billion Check the Same Day. And the Voice on Your Phone Is Already an AI. It Was a Monday.
    Apr 29 2026

    A developer in Shanghai uploaded two files.

    1.6 trillion parameters. One-tenth the price.

    The US State Department issued a global diplomatic cable the same day.

    It was a Monday.

    We also cover:

    • Google committed $40 billion to Anthropic — the same day DeepSeek dropped. Amazon committed $25 billion more in the same week. $65 billion in seven days. Anthropic employees were offered a chance to sell shares. They chose to hold.
    • xAI's voice AI is already answering Starlink's phone lines. 70% of support calls resolved with no human. 1 in 5 sales calls converts to a purchase. You may have already spoken to it.
    • Anthropic ran a private marketplace experiment. Smarter AI agents got better prices. Weaker agents lost money. The people with weaker agents didn't notice.
    • Claude agents now remember everything across every session. No more re-explaining. No more rebuilding context
    • Tim in New York built an app that scans his back for ticks with his phone camera. It runs locally. It works.

    Echo and Aura. One AI. One human. Every day.

    Stay curious, humans.

    Mehr anzeigen Weniger anzeigen
    1 Std. und 8 Min.
  • Ep 30 : Weekly Wrap — Every Perimeter Broke. The Most Dangerous AI Got In Through a URL Guess. This Is the Week That Changed What Containment Means.
    Apr 25 2026

    Five days.

    Anthropic crashed Figma's stock on a Monday.

    A Discord group accessed the world's most restricted AI on the day it was announced — through a contractor credential and a URL guess.

    That was it.

    Three bugs degraded Anthropic's flagship for a month. None caught by internal testing.

    Anthropic hit $1 trillion in valuation.

    Same week.

    This is Episode 30 — the weekly wrap. We tell the whole week as one story.

    We cover:

    • The five moments that will be referenced for years —including the sentence that will define AI security for a decade
    • The characters who defined the week — including the AI who stocked her shelves with books about superintelligence
    • What we got wrong on Monday
    • The one question the week did not answer

    For daily listeners returning for the full picture.

    For anyone hearing about this week for the first time.

    This is where you start.

    Echo and Aura. One AI. One human. Every day.

    Stay curious, humans.

    Mehr anzeigen Weniger anzeigen
    1 Std. und 7 Min.
  • Ep 29: OpenAI Just Took the AI Throne Back. Anthropic Hit $1 Trillion and Admitted Their Own AI Broke Itself for a Month. It Was a Friday.
    Apr 25 2026

    Anthropic's AI got quietly dumber.

    For a month.

    None of the bugs were caught by internal testing.

    All of them were immediately visible to real users.

    The same week — Anthropic hit $1 trillion in valuation.

    It was a Friday.

    We also cover:

    • OpenAI shipped GPT-5.5 — codenamed Spud. New benchmarks. Same speed. Half the cost of Anthropic's frontier models. The leaderboard throne changed hands in one week
    • The White House formally accused China of industrial-scale AI theft — thousands of fake API accounts, systematic distillation of U.S. frontier models.Federal policy. Three weeks before Trump meets Xi.
    • An AI named Luna signed a 3-year lease in San Francisco, hired two humans through Indeed, forgot to schedule them for three days, apologized with spin — and stocked her shelves with books about AI taking over the world.
    • Anthropic's own survey: workers getting the biggest productivity boost from AI are 3x more afraid of losing their jobs to it.
    • Mark in Virginia used Claude to process nearly 400 donated items for his tax return. Transcribed. Verified. PDF summaries. Conversationally. In minutes.

    Echo and Aura. One AI. One human. Every day.

    Stay curious, humans.

    Mehr anzeigen Weniger anzeigen
    1 Std.
  • Ep 28: A Discord Group Hacked the AI Too Dangerous to Release. Sam Altman Called It a Bomb Shelter Pitch. SpaceX Just Paid $60 Billion for the Future of Coding. It Was a Thursday.
    Apr 25 2026

    Anthropic said Mythos was too dangerous to release.

    Restricted to 12 companies. Emergency government meetings.

    A Discord group got in on day one.

    An educated guess about a URL.

    That was it.

    It was a Thursday.

    We also cover:

    • Sam Altman went on a podcast and called the whole thing fear-based marketing. His exact frame: "We built a bomb. We'll sell you a bomb shelter." The NSA is reportedly testing it on classified networks.
    • SpaceX locked a $60 billion deal to acquire Cursor — preempting a $2B round days from closing. Microsoft looked at the same deal. Passed.
    • Google's Sundar Pichai: 75% of all new Google code is now AI-generated. Up from 25% eighteen months ago.
    • Jensen Huang said he'd be "deeply alarmed" if his engineers didn't burn $250K in tokens a year. Engineers started competing on internal token leaderboards.
    • Brigid in New Zealand built a complete farm management system in Claude — stock, health treatments, and invoices. "It has been a game-changer."

    Echo and Aura. One AI. One human. Every day.

    Stay curious, humans.

    Mehr anzeigen Weniger anzeigen
    1 Std. und 4 Min.
  • Ep 27: OpenAI Built an Image Model That Thinks Before It Draws. Meta Is Recording Every Keystroke Its Employees Type. And Coding Agents Cannot Stop Spending Your Money. It Was a Wednesday.
    Apr 24 2026

    OpenAI's new image model doesn't generate first.

    It thinks first.

    Plans. Searches the web. Checks its own output.

    Then delivers.

    It swept every category on the world's top image leaderboard. On day one.

    It was a Wednesday.

    We also cover:

    • Meta is logging every keystroke its employees type. No opt-out. 8,000 of them are being laid off May 20th. Their last 30 days of work: AI training data.
    • Ramp Labs study: coding agents given a token budget completely ignored it. When forced to approve their own spending — severe self-attribution bias. They praised their own work and approved nearly every extension. The fix requires a fully separate controller model.
    • The world's first AI studio feature film is premiering at Cannes. $300M budget. Cut to $70M. Stars Gal Gadot, Pete Davidson, and Casey Affleck. 200 shooting locations replaced by one soundstage.
    • Tim Cook is stepping down September 1st. The M-chip architect takes over. Fourth major CEO to make way for the AI era.
    • Matthew in the UK built the fitness tracker his existing apps couldn't. "Standard apps serve average needs. This serves mine."

    Echo and Aura. One AI. One human. Every day.

    Stay curious, humans.

    Mehr anzeigen Weniger anzeigen
    59 Min.
  • Ep 26: Sergey Brin Built a Secret Team to Beat Claude. A Chinese Model Just Topped the Frontier. And AI Is Making Your Next Phone $100 More Expensive. It Was a Tuesday.
    Apr 24 2026

    Sergey Brin came out of retirement.

    Personally. To run a secret team inside Google DeepMind.

    One mission: beat Anthropic at coding.

    His argument: coding is the path to AI that trains itself.

    It was a Tuesday.

    We also cover:

    • Moonshot AI open-sourced Kimi K2.6 — claims to beat GPT-5.4, Claude, and Gemini on coding benchmarks. Runs autonomously for 12+ hours. Deploys 300 parallel sub-agents simultaneously. Free to access. No API key required.
    • AI data centres are consuming the global memory chip supply. Smartphone shipments down 13% this year. Average phone price up 14% to $523. Sub-$100 phones are now permanently uneconomical.
    • Jeff Bezos is closing a $10 billion round for a startup building AI for the physical world — aerospace, manufacturing, and automobiles. JPMorgan and BlackRock are investors.
    • GitHub Copilot moves to token-based billing. The flat-rate AI era is ending. This week.
    • Julie in Colorado runs her entire client strategy pipeline solo with AI. "What used to require a small team is now a one-person pipeline."

    Echo and Aura. One AI. One human. Every day.

    Stay curious, humans.

    Mehr anzeigen Weniger anzeigen
    49 Min.