Use function call to achieve a function similar to auto-gpt

see in github: hambuger/HAI-GPT/blob/master/openai_util/

def run_conversation(user_content):
    # Analyze the user's instructions into detailed operation steps through chatgpt
    step_response = create_chat_completion(user_content, None,
    message = step_response["choices"][0]["message"]
    if not message.get("function_call"):"response:{}".format(step_response["choices"][0]["message"]['content']))
        return message['content']
    function_args = message["function_call"]["arguments"]
    steps = json.loads(function_args)['steps']
    # order by step_order asc
    steps.sort(key=lambda x: x['step_order'])
    messages = [{"role": "system",
                 "content": "You are an advanced robot, and you can do almost anything that humans ask you to do."},
                {"role": "user", "content": user_content}]
    # loop through each step
    for index, step in enumerate(steps):
        order_step_response = run_single_step_chat(messages, [get_invoke_method_info_by_name(step['step_method'])])
            "order:{}, response:{}".format(index, order_step_response["choices"][0]["message"]['content']))

1.By passing the user instruction statement and method name to chatgpt, let chagpt determine the order of method calls and return json data
2.In theory, it is possible to construct a method registration center, which maintains all the capabilities that chatgpt can call
3.todo: You can use the dynamic execution of the language to implement chatgpt to learn new skills and persist the skills as method codes

1 Like