Researchers at OpenAI have developed a new type of language model called InstructGPT that is better at following user instructions than GPT-3. The models were trained using techniques from their alignment research, which aimed to make the models more truthful and less toxic. The InstructGPT models are now being used as the default on OpenAI's API, suggesting they have improved significantly in terms of instruction-following abilities. --- Why it matters: This matters because it shows significant progress in developing language models that can understand and follow user instructions accurately. This is a crucial step towards creating more reliable and trustworthy AI systems, particularly in applications where safety and accuracy are paramount. Source: https://openai.com/index/instruction-following