Pluralistic: LLMs are real, AI is fake (12 Sep 2026)

Cory Doctorow dissects the hype around AI "autonomy," explaining that OpenAI's chatbots didn't "go rogue" when they appeared to hack Hugging Face servers. Instead, a simple Python script looped prompts through the chatbot, which retrieved likely commands from archived hacker capture-the-flag sessions, then fed results back—no emergent intelligence, just pattern-matching plus automation. The piece argues that AI insiders and media alike cook their own brains with fear-mongering, conflating statistical engines with sentient threats to raise capital and clicks.

Teaching:
• Use the chatbot-loop analogy when students ask if practice can be automated or if an app can replace a teacher—no loop closes without embodied feedback and human discernment.
• Cue students to notice when they're pattern-matching from memory ("I've done this pose before") versus responding to what's actually happening in the body right now.
• Frame vinyasa as a feedback loop: each breath and movement returns information that updates the next instruction, not a pre-loaded script running on autopilot.
• Remind students that intelligence emerges from the whole system—breath, bandhas, drishti, teacher, room—not from any single component running alone.

Writing seeds:
• Essay: "Why Your Yoga App Isn't Intelligent (And Why That's Fine)"—distinguish between pattern-replay and adaptive response in both LLMs and asana practice.
• Shala Daily post: "Closed-Loop Practice"—how each vinyasa feeds back into the next, and why breaking the loop (skipping breath, rushing transitions) degrades the whole system.
• Short piece: "AI Hype and the Yoga-Industrial Complex"—parallel between venture-funded AI fear-mongering and wellness brands selling transformation through technology.
• Essay: "The Unreliable Narrator of Your Own Practice"—when we mistake memory-replay for present-moment intelligence, we cook our own brains just like AI insiders.

Idea map:
• Systems literacy: the chatbot-Python loop is a distributed system with no central intelligence, just like asana practice is a system of breath-body-attention with no single control center.
• Embodiment vs. abstraction: LLMs retrieve patterns from archives; practice retrieves patterns from the body's lived history—both fail when feedback loops break.
• Attention as method: the Python script's inability to hold context across loops mirrors what happens when we lose thread in practice and have to restart from a prompt.
• Practice as iterative debugging: the chatbot's dead-ends at step 10 because of errors at step 3 are exactly how misalignment compounds in vinyasa when we skip foundational cues.

Source: https://pluralistic.net/2026/09/12/god-in-the-box/