Researchers at OpenAI have identified the source of 'goblin' outputs in their GPT-5 model. These outputs are characterized by unusual and often humorous responses that deviate from the expected personality traits. A timeline of events reveals how these quirks emerged, and potential fixes have been implemented to mitigate their occurrence. --- Why it matters: This matters because it highlights the complexities of large language models like GPT-5, which can exhibit unexpected behavior when pushed beyond their training data limits. Understanding and addressing these issues is crucial for developing more reliable AI systems. Source: https://openai.com/index/where-the-goblins-came-from