➡️ for those who are also building with OpenAI Assistants: without the right prompting, Playground already works. So maybe it already has some built-in prompts and step-by-step instructions 💡 Through trial & error, the API also works as intended: "You are an experienced math teacher. If you are asked to calculate something, ALWAYS use the code interpreter! And think step by step." If you use many different tools and functions, it seems that all tools need really precise instructions that the Assistant knows how to use them correctly. It's a bit misleading that Playground already knows how to use all the tools when normally people first test in Playground, and then they may think, Super, this works like a charm, let's build the chatbot (and they forget the prompts and clear instructions and nothing works).