OpenAI Codex and Goblins
Recently, you might have read about how ChatGPT started talking about goblins, gremlins and pigeons more and more in its user responses. For instance, instead of saying something like “there’s a bug in the code”, it would say “there’s a program goblin present”. OpenAI’s response to this was somewhat indicative of how difficult it is to control LLM’s in what they can potentially say to a user, and that was by adding a line to the codex-rs/models-manager/models.json personality, basically telling the model to: ...