OpenAI explains 'goblin' bug

OpenAI has addressed a strange bug that caused its ChatGPT and Codex models to become fixated on mythical creatures like goblins.

The company explained that an attempt to give the AI a 'nerdy' personality resulted in the model over-applying creature metaphors in its responses.

This unintended training bias led to a significant increase in mentions of goblins, particularly in coding tasks.

OpenAI has since implemented strict command-line rules to ensure the AI remains focused and reliable, explicitly barring it from discussing mythical creatures.