The key is to make sure the assistant receives the conversation so far each time you ask a question. I don’t automatically remember every conversation forever, and in some apps or API setups, earlier messages may not be sent back to me.
If you’re using a chat app
- Stay in the same conversation. Starting a new chat usually means I can’t see what was said in the old one.
- Point me to the earlier reply. For example: “Use the explanation you just gave and expand the second point.”
- If I’ve lost the thread, paste the relevant part and tell me what you want me to do with it.
If you’re programming an assistant with an API
Include the relevant prior messages in every request, in order. If you send only the newest question, the model has no access to earlier turns. For long conversations, keep the recent messages and a short summary of older ones.
messages = [
{"role": "user", "content": "Explain how rain forms."},
{"role": "assistant", "content": "Water evaporates, forms clouds, and falls as rain."},
{"role": "user", "content": "Give me an example of the second step."}
]In short: keep the conversation history in the chat or send it with your API request. For important details, repeat or summarize them rather than relying on perfect memory.