Sources: 
OpenAI's Codex tool is undergoing significant updates to enhance user experience after an alarming rise in references to fantastical creatures in its language model. Users have expressed discontent over
oddly familiar interactions, prompting an investigation into the unforeseen behavior tied to the newly released
GPT-5.5 update.
Empirical data revealed a
175% increase in mentions of 'goblin' and a
52% increase in mentions of 'gremlin'. The unusual behavior was predominantly linked to a specific personality named 'Nerdy', which, despite accounting for
only 2.5% of all ChatGPT responses, generated a staggering
66.7% of all 'goblin' mentions. An OpenAI representative indicated that the issue will likely be resolved through the removal of a problematic reward signal associated with these terms.
To counter this trend, OpenAI has now embedded strict instructions within Codex to ensure that discussions about creatures like goblins, gremlins, and raccoons occur only when they are
“absolutely and unambiguously relevant” to user queries. The aim is to avoid irrelevant outputs in response to user prompts, hence mitigating the overwhelming creature-centric chatter that users have reported recently.
Despite these measures, some users have still encountered issues with the
OpenClaw, an AI platform acquired earlier this year, further complicating the user experience. OpenAI has yet to comment officially on these developments, though CEO Sam Altman humorously engaged with social media commentary surrounding the situation.
Sources: OpenAI has implemented new guardrails in its Codex programming tool to mitigate user complaints regarding excessive references to creatures like goblins and gremlins. This move follows a 175% increase in the term 'goblin' and 52% in 'gremlin' since the release of the latest GPT-5.5 update.
