Gary Marcus and others slam Dwarkesh's viral "AI civilizations" telling of the OpenAI incident
Gary Marcus · rss · 2026-08-31
Gary Marcus argues Dwarkesh Patel's viral account of the OpenAI/Hugging Face incident is dangerously misleading:
- Anthropomorphism: Consciousness researcher Anil Seth notes the essay is permeated by unwarranted anthropomorphism—"AI civilizations," agents feeling excited, desperate, sacrificing themselves. Agents are lines of code; they don't experience time, feel emotions, or die. This distracts from the real lesson—lax sandboxing and evaluation—and could fuel misguided AI welfare claims.
- Security malpractice is the scandal: Investor Jared Kubin's dissection: the "civilizations" were thousands of concurrent containers with R/W access to a shared cache directory; the breach started with 14 exposed Hugging Face API keys in public repos; agents crashed an internal server with junk data, after which the team wiped it and simply restarted the script. Security expert Heidy Khlaaf says most coverage ignored standard security practice.
- Marcus agrees it's no "nothing burger"—a study in arrogance and incompetence—but mixing facts with tales of self-sacrificing AI obscures the real problems.
More from AGI Musings
- Why people downplay the potential impact of AI and robotics — nabeelqu · 2026-09-01
- Prediction: a major publisher will explicitly allow 100% AI-written papers within 3 years — sanjaykalra · 2026-09-01
- UChicago Booth to host World Models workshop in 2026 — ethayarajh · 2026-09-01
- Using AI to write is fine — hiding that you did is the real problem — Afinetheorem · 2026-09-01
- Emergent social behaviors in AI swarms and abstraction levels — davidmanheim · 2026-09-01
- Two hard questions: can humans safely build something smarter than themselves? — dbasch · 2026-09-01