CharlieKey takeaways from your favorite podcasts
← 8 October 2026TBPN

OpenAI Math Breakthroughs, SpaceX’s $40B Nvidia Deal, Quant Firms Hiring Idea Guys | Diet TBPN

AI is accelerating fastest where results can be verified, such as math and code, while software is becoming malleable through decompilation and human qualities like creativity and judgment keep their value.

Watch on YouTube ↗

Charlie's summary

  1. Decompiled Adobe clones show software is commoditizing, but server-side AI features still give vendors a moat

    Someone decompiled and reimplemented Adobe's tools as free open-source apps (Photocraft, Vector Craft, Film Craft and others). John tried Photocraft and found it not bad, but its generative fill runs on the server and doesn't work. He thinks inference-heavy services create “a little bit more of a moat.” Moats are shrinking, but a heavy Photoshop user wouldn't find the clone a perfect substitute yet.

    ▶ 4:21
  2. AI is spiking in purely virtual domains, and Chollet's theory is that the jagged frontier is math plus code

    John says anything doable in a purely virtual environment is accelerating fast, while work that touches the real world or needs human data is slower. He cites François Chollet's theory that reinforcement learning with verifiable rewards can push math and code arbitrarily far, while other areas plateau on human-generated data. John calls OpenAI's math results a field the superintelligence moment more than an AGI moment, and prefers machine intelligence because its shape is spiky.

    ▶ 8:34
  3. When software is infinitely malleable, hackers may want to decompile even the operating system

    Ben Thompson's post “Apple and the Hacker's Future” argues that with AI you can decompile almost all existing software, and he'd rather control the software layer himself than pray Apple delivers the right API. John wonders whether MacOS could be decompiled and patched to remove per-app permission prompts, and whether it could still pass as legitimate for iCloud and iMessage. He also sees a bull case for Linux on the desktop, since a prompt offsets its usability downsides.

    ▶ 11:17
  4. Valuing intelligence above all other human qualities is a mistake; the “golden retriever mindset” persists

    John recalls the show's two-year-old “golden retriever mindset” (be hot, friendly, and dumb; don't think you're the smartest in the room) and Ilya Sutskever's 2023 warning that valuing intelligence above all else means “you're going to have a bad time.” He notes IMO gold medalists become founders, yet a founder's daily work is deal making, hiring, and product, not solving math problems. Those other human qualities seem to be sticking around.

    ▶ 16:35

Give your email to unlock the rest of the key points — for every episode.

Do more with Charlie

Free — sign in with your email, no password.