create conversations without —continue

[skybrian]

Unless I missed something, it looks like conversations in llm are always constructed with the —continue flag.

I’m writing a vscode plugin that allows users to chat with an LLM and edit the chat history, which is stored in a text document or notebook. It would be nice to be able to construct a conversation entirely outside llm and send it in on stdin, perhaps using a JSON format.

A workaround is to send the conversation in anyway, as a single prompt, using any format you like. GPT4 seems to understand it well enough. But I wonder what the difference would be if it were sent using GPT4’s normal format for conversations?