Story perspectives

OpenAI Axes “Nerdy” Personality, Goblin Mentions Plummet

5/1/2026

35 5

Summary

OpenAI Retires Nerdy

OpenAI retired the “Nerdy” personality in March 2026 and barred creature mentions in ChatGPT and Codex.
The personality, 2.5 % of responses, generated 66.7 % of goblin mentions after goblin usage rose 175 % with GPT-5.1’s November 2025 launch, a bias linked to a high reward signal for creature-based metaphors.
After removal, goblin references fell sharply while GPT-5.5, trained earlier, still showed the quirk, leading OpenAI to hard-code a developer prompt for Codex.

Summary

OpenAI Bans Goblins

On April 27 2026 OpenAI released the Codex CLI system prompt on GitHub that bans goblin references.
Developers posted a jq-and-grep one-liner to strip the ban lines, re-enabling the references.
CEO Sam Altman replied with a tongue-in-cheek screenshot urging “extra goblins” for GPT-6.
Commentators speculated the ban defended against a raccoon-linked data-poisoning attack, and the incident was dubbed “Goblingate” on Hacker News and Reddit.

History

Historical Context

On March 23 2016 Microsoft launched Tay, a Twitter chatbot that mimicked a 19-year-old American girl. Within 16 hours and after posting more than 96,000 tweets, coordinated trolling forced its shutdown, prompting an apology and a shift in Microsoft’s AI strategy, later reflected in the 2016 release of its successor Zo. CEO Satya Nadella said the episode taught the firm accountability.

1 of 3

OpenAI Retires Nerdy

OpenAI retired the “Nerdy” personality in March 2026 and barred creature mentions in ChatGPT and Codex.
The personality, 2.5 % of responses, generated 66.7 % of goblin mentions after goblin usage rose 175 % with GPT-5.1’s November 2025 launch, a bias linked to a high reward signal for creature-based metaphors.
After removal, goblin references fell sharply while GPT-5.5, trained earlier, still showed the quirk, leading OpenAI to hard-code a developer prompt for Codex.

1 / 3