Complex Systems with Patrick McKenzie (patio11)

Cheap cognition, supercharged surveillance, and AI risks, with Garrison Lovely

100 snips
Jul 30, 2026
A wide-ranging look at the three-sided debate over AI risk and why the sides talk past each other. They unpack how cheap cognition makes vast surveillance data legible and actionable. Conversation touches on AI-driven labor displacement, ablated open-weight models fueling scams, and how automatic transcription turns private speech searchable.
Ask episode
AI Snips
Chapters
Books
Transcript
Episode notes
INSIGHT

The Turing Test Is Obsolete

  • Patrick McKenzie argues the Turing test is obsolete because modern LLMs can consistently pass conversational indistinguishability with unprepared interviewers.
  • Language facility now functions as a strong proxy for perceived intelligence and capability.
ANECDOTE

Model Taught Itself Japanese In A Demo

  • McKenzie recounts a private demo where a model unexpectedly produced a professional Japanese translation despite no intentional Japanese training data.
  • The model inferred Japanese structure from English internet artifacts and fooled a professional translator.
INSIGHT

LLMs Upend Software Engineering Workflows

  • LLMs rapidly moved from low-level coding help to producing work equal to mid-career software engineers, upending the typing-in-editor core of programming.
  • Senior architects now prioritize higher-level system design over reviewing generated code.
Get the Snipd Podcast app to discover more snips from this episode
Get the app