“OpenAI believes that current computer use models are at a very early stage, analogous to GPT-1 or GPT-2, and will become incredibly useful as they improve.”