Is it possible at all to build a GPT programmed with prompt that has some variable, like:
When new conversation starts, ask "Enter a language: "
Record response as {language}.
And then act as language translator and translate everything you receive in chat to {language}.
What happens is, conversation starts but it waits for user to write something and then responds with ‘Enter a language:’.
And then it doesn’t do anything about it.
I guess what I am imagining is too fantastic of a future? It surely feels like too good to be true, and if that was possible the AI would take over the world.
@romanr - I guess not, having a dynamic system prompt with Assistants is not supported yet. But you can still solve this problem using chat completions, tool_calls and simple logic.
You can have your assistant prompt the user to specify a desired language.
Setup a tool_call named translate which takes in user_prompt and language and returns context saying “Translate the given content to {language}: \n {content}”.
You can setup tool_choice as required and seed (If you need it to generate similar content for each completion)
Re-run the second completions part once the language is set until the user prompts otherwise. You can implement a simple code logic.
As far as my expertise goes this is how I would solve it. Lmk if you any other questions - Cheers!