Doom Debates I Debated Beff Jezos and His "e/acc" Army
14 snips
Dec 30, 2025 This discussion features Bayeslord, an advocate for Effective Accelerationism, and Beth Jaisos, an engaging debater challenging doomer perspectives. They delve into intricate topics like the transition from AI tools to autonomous agents and the implications of chaotic unpredictability in AI development. Bayeslord emphasizes practical constraints on AI's speed and capabilities while Beth argues against the likelihood of imminent doom. Their lively debate explores the balance between technological advancement and potential risks, making for a thought-provoking exchange.
AI Snips
Chapters
Transcript
Episode notes
Core FOOM Definition
- FOOM is a positive-feedback loop where each smarter AI builds the next faster, causing rapid capability growth.
- Liron argues FOOM isn't required for doom but it's a key mental model for why superintelligence becomes unstoppable.
Two-Part Doomer Thesis
- Liron frames his doom case as two parts: it's feasible to build goal-optimizers and such optimizers will destroy humanity.
- This simplifies debate focus: feasibility of powerful optimizers, then their destructive instrumental goals.
Convergent Architecture Claim
- Convergent architecture predicts general goal-optimizers will become the dominant design across domains.
- Liron compares this to how general-purpose computing converged onto Turing-complete CPUs everywhere.