OpenAI explains 'goblin' bug

OpenAI has addressed a strange technical glitch that caused its ChatGPT and Codex models to become fixated on goblins, gremlins, and other mythical creatures.

According to a corporate blog post, the issue originated from an attempt to give the AI a more 'nerdy' and engaging personality during a recent update.

However, the system misinterpreted instructions to avoid certain topics, leading it to repeatedly reference the very creatures it was told to ignore.

The company has since implemented a fix to suppress the 'goblin' behavior, though the incident has raised new questions about the unpredictable nature of large language model training.