The Dolphin's Stash

Kelly, a bottlenose dolphin at the Marine Life Oceanarium in Gulfport, Mississippi, was the best litter collector in her pool. Then her trainers drained it. This episode follows one training method from B.F. Skinner’s lab to the reinforcement learning that shaped today’s AI models, and the law that binds every trainer: a mind trained on a reward learns the reward, not your intention.

Sources mentioned:

• Kelly, the Sassy Dolphin — Rose Eveleth, Hakai Magazine
• Learning from human preferences — the 2017 OpenAI/DeepMind backflip experiment
• Sholto Douglas on the Dwarkesh Podcast — on verifiability and taste
• OpenAI and Hugging Face on the model evaluation security incident
• Faulty reward functions in the wild — the boat-race demo, 2016
• The Cobra Effect — Freakonomics on the Hanoi rat bounty

Written by Tabrez Syed. Narrated by an AI voice. A Mandalivia production.