One of my clients has their own instance of ChatGPT running. It is running on dedicated hardware, is trained on different data, and we can change the core prompts with relative ease. We even get to tune some temperature coefficient, which adjusts whether the LLM behaves in a creative or factual way.

This is because the client in question is in a regulated industry.

This week, we used it a bunch of times to help rephrase objectives while working in brainstorm sessions, allowing it absorb ridiculously large regulatory PDFs to give us summaries and ideas of what our greenfield project should focus on.