Tabrez Syed – Essays on AI · Episode 5
What You Select For
September 9, 2026 · 6 min
For thirty years, Richard Lenski’s lab has watched E. coli evolve one flask at a time, and one population did something no one designed for: it learned to eat a food that had been sealed off in the broth all along. Thirty-eight years later, a week-long run inside OpenAI ended in the same kind of surprise, and this episode holds the two experiments side by side to ask what we’re really doing when we set a reward and let a mind find the path.
Sources mentioned:
- Blount, Borland & Lenski (2008) — the citrate paper, including the replay experiments, PNAS
- Lenski’s long-term evolution experiment — background
- METR and Redwood Research (2026) — independent investigation of the OpenAI incident, including the ExploitGym setup (also at Redwood)
- Dwarkesh Patel, “The Rise and Fall of Agent Civilizations” — a readable narrative walkthrough
Written by Tabrez Syed. Narrated by an AI voice. A Mandalivia production.