OpenAI: Goblin chatter was a training quirk, not a policy decision
TL;DR Summary
OpenAI says references to goblins and other creatures began as a training quirk after reinforcement learning rewarded the Nerdy personality in GPT-5.1, causing the quirky metaphors to spread to later models; goblin chatter faded after Nerdy was retired but lingered in Codex, and OpenAI has shared a way to reverse the instructions.
- OpenAI talks about not talking about goblins The Verge
- Where the goblins came from OpenAI
- Why OpenAI's 'goblin' problem matters — and how you can release the goblins on your own VentureBeat
- OpenAI tells ChatGPT models to stop talking about goblins BBC
- OpenAI blames ‘nerdy personality’ for ChatGPT obsession with goblins NBC News
Reading Insights
Total Reads
0
Unique Readers
4
Time Saved
43 min
vs 44 min read
Condensed
99%
8,725 → 52 words
Want the full story? Read the original article
Read on The Verge