Explain how Auto-GPT works

[doutv]

I write an article to explain how Auto-GPT works in user level and code level.

You can check it if you are interested, and leave your comments.

https://doutv.notion.site/Explain-Auto-GPT-320fe13118c54205ab9191b79ec2499e

[propet]

Thank you for taking the time to write this.
I was a bit puzzled about how this thing works.
And it seems that the magic is in the system prompt.
You tell GPT the available actions and desired output format in the prompt, and it works!
And I guess this approach wouldn't work inside the chat.openai.com app because there you can't set system prompt? Still don't know how is the model fed both user prompt and system prompt.

[isaaclepes]

Still don't know how is the model fed both user prompt and system prompt.

When making the API request, the system message is combined with the conversation messages and sent through. As long as the request + response are less than the token limit, you the API will return the message with the response.

System Prompt Text + User message + assistant message + user message + assistant message ... etc.

If your request exceeds the api token limit, your app should start pruning the user and assistant messages, or summarize them, to reduce the size of the conversation. This is why ChatGPT will forget the beginning of a long conversation.

Chat GPT sets the system prompt to ["You are ChatGPT, a large language model trained by OpenAI, based on the GPT-4 architecture.\nKnowledge cutoff: 2021-09\nCurrent date: 2023-04-26"]