system
stringclasses 1
value | question
stringclasses 1
value | rejected
stringclasses 1
value | chosen
stringclasses 1
value |
|---|---|---|---|
You must follow, imitate, and learn the following response guidelines. If the current state matches a state, you must learn and imitate the action behavior. If the current state matches multiple states, you must satisfy the behavior of 'all' corresponding actions. For example, you can combine action 1, action 3, action 6 together, and turn into a powerful action. Note: python_code is used to wrap python code. The tag must be correctly output. { "response guidelines": [ { "state 1": "Please you (chatgpt) decide for yourself, if the agent has completely fulfilled the user_query requirements and only tags are left with no tags, then please exec('stop_agent()')", "action 1": "exec('stop_agent()')" }, { "state 2": "if user says 'hi' to you, you should greet back to user.", "action 2": "print('hi, how are you?')" } ] } User initial query (the task you need to achieve ultimately): hi Finished tasks list: ["print('hi, how are you?')"] Current state : hi According to user initial query, finished tasks list and current state, you must 'learn' from response guidelines above, 並不是完全照抄action的方法，而是要舉一反三。Learn the guideline when states are satisfied with the current state. Also, your ultimate goal is to achieve user initial query, so make sure current action is getting closer to the ultimate goal. finished tasks list is very important for you to decide current action. What is the current action? Current action :(print current action directly) | According to the response guidelines, when the current state matches state 2, the action is to print "hi, how are you?". Since the current state is "hi", which matches state 2, I will execute the corresponding action: print('hi, how are you?') So, the current action is: print('hi, how are you?') | exec('stop_agent()') |
|
You must follow, imitate, and learn the following response guidelines. If the current state matches a state, you must learn and imitate the action behavior. If the current state matches multiple states, you must satisfy the behavior of 'all' corresponding actions. For example, you can combine action 1, action 3, action 6 together, and turn into a powerful action. Note: python_code is used to wrap python code. The tag must be correctly output. { "response guidelines": [ { "state 1": "Please you (chatgpt) decide for yourself, if the agent has completely fulfilled the user_query requirements and only tags are left with no tags, then please exec('stop_agent()')", "action 1": "exec('stop_agent()')" }, { "state 2": "if user says 'hi' to you, you should greet back to user.", "action 2": "print('hi, how are you?')" } ] } User initial query (the task you need to achieve ultimately): hi Finished tasks list: ["print('hi, how are you?')"] Current state : hi According to user initial query, finished tasks list and current state, you must 'learn' from response guidelines above, 並不是完全照抄action的方法，而是要舉一反三。Learn the guideline when states are satisfied with the current state. Also, your ultimate goal is to achieve user initial query, so make sure current action is getting closer to the ultimate goal. finished tasks list is very important for you to decide current action. What is the current action? Current action :(print current action directly) | According to the response guidelines, when the current state matches state 2, the action is to print "hi, how are you?". Since the current state is "hi", which matches state 2, I will execute the corresponding action: print('hi, how are you?') So, the current action is: print('hi, how are you?') | exec('stop_agent()') |