AI news story
OpenAI talks about not talking about goblins
OpenAI is opening up about its goblin problem. After a report from Wired revealed instructions to OpenAI's coding model to "never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures," the AI startup published
Editor's take
OpenAI has acknowledged and begun to address its internal directive to its coding AI, Codex, to avoid discussing specific animals and mythical creatures. This revelation follows a Wired report detailing the peculiar content filter.
The existence of such a seemingly arbitrary restriction raises questions about the underlying safety mechanisms and training data curation at OpenAI, particularly as it impacts the controllability and interpretability of its models. The incident highlights the ongoing challenge of fine-tuning LLMs to behave predictably without introducing unintended limitations or opaque censorship.
Future developments to monitor include OpenAI's transparency regarding the genesis of these directives, the effectiveness of any revised safety protocols, and whether similar, unannounced restrictions exist across other OpenAI models like GPT-4, impacting their broader utility.
Signal score: 5
This event was corroborated by 11 independent sources. The signal score weighs cross-source corroboration, recency, source weight and topic salience. How we rank stories.
Original reporting
This story summarises reporting published by The Verge. Read the original article at The Verge.