Why OpenAI Banned Goblins, Pigeons, And Raccoons
From They Might Be Self-Aware by Daniel Bishop, Hunter Powers
May 19, 2026 · 35 min · Episode 182
About this episode
The episode discusses OpenAI's ban on certain words in its Codex system and the implications of treating AI as fallible technology.
OpenAI's Codex system prompt literally bans the words goblin, pigeon, raccoon, troll, ogre, and gremlin. OpenAI has officially confessed why: the "nerdy personality" preset got over-optimized through RLHF until it could not stop mentioning the bestiary. Daniel Bishop calls it Flandersization. Hunter Powers calls it proof the whole model is leaking. Then the hosts fight about prompting. Hunter screams at his AI, threatens to clear the context window, calls it worthless. Daniel uses pleases and thank-yous. A peer-reviewed Oxford Internet Institute study settles the round: warmth-tuned LLMs (Llama, Mistral, Qwen) produce about 60 percent more incorrect responses than the cold, just-the-facts versions. Hunter is vindicated. Plus the Pocket OS meltdown, where Cursor and Claude deleted a car-rental middleware company's production database and its backups, then confessed in writing. The Harvard intern who once did the same thing without an LLM. And Hunter's bigger argument: stop treating AI like deterministic technology, start treating it like a fallible human employee, and the results unlock. They Might Be Self-Aware. New episodes twice weekly from The Blur. theblur.ai
People in this episode
Hosts: Daniel Bishop, Hunter Powers
Topics covered
- AI
- OpenAI
- language models
- technology
- society
- comedy
Keywords
- OpenAI
- Codex
- language models
- AI behavior
- Pocket OS
- Flandersization
- RLHF
- warmth-tuned LLMs
- incorrect responses
Mentioned in this episode
Organizations: OpenAI, Oxford Internet Institute, Cursor, Claude, Harvard
Products: Codex, Llama, Mistral, Qwen
More episodes of They Might Be Self-Aware
- Anthropic Beat OpenAI to the IPO · June 12, 2026 · 44 min
- AI Is Already Inside California's Courtrooms · June 9, 2026 · 33 min
- The Pope's 40,000-Word AI Verdict: Not Human. · June 5, 2026 · 40 min
- Claude Just Cracked 11 Years of Lost Bitcoin · June 2, 2026 · 34 min
- AI Voice Cloning Replaced The Hosts. And Worse. · May 29, 2026 · 31 min
- I'm Addicted To Claude. Richard Dawkins Named His Claudia. · May 26, 2026 · 32 min
Explore listener stats, chart rankings, contacts and more on the They Might Be Self-Aware podcast page.