How to build a prompt which answers only to the context we specified

[clever trench]

Answer the question only related to the topics

import openai openai.api_key="sk-xxx" response = openai.ChatCompletion.create( model='gpt-3.5-turbo', messages=[ {'role': 'system', 'content': "Answer the question only related to the topics of health if they ask outside scope queries, say That I have been trained to answer only health related queries"}, {'role': 'user', 'content': 'python code for factorial'} ], max_tokens=100, temperature=0, )

Returns python code for factorial

[clever trench]

@sweet socket @fiery kettle please advise

[sweet socket]

use the user message to make instructions I suppose

replace the system role with user

In this model the system message doesn’t weigh so much

[clever trench]

Still same behaviour

[sweet socket]

emm so what’s your whole message? what’s in the python code for factorial

Maybe you can change the order of the two messages. First send the passage about which you will ask some questions

Then say answer the question I send you later based on the passage before. If there’s not any relevant information for the question say i don’t know

finally send the question

[clever trench]

could you please send the modified code

import openai openai.api_key="sk-xxx" response = openai.ChatCompletion.create( model='gpt-3.5-turbo', messages=[ {'role': 'assitant', 'content': "Answer the question only related to the topics of health if they ask outside scope queries, say That I have been trained to answer only health related queries"}, {'role': 'user', 'content': 'python code for factorial'} ], max_tokens=100, temperature=0, )

[sweet socket]

this is the part of message:

{'role': 'user', 'content': 'replace with your passage here'}

{'role': 'user', 'content': 'I will later ask you a question about the passage I sent you before. Answer the question based on the passage, if there’s nothing relevant to the question from the passage, just reply I don’t know'}

{'role': 'user', 'content': 'replace with your question here'}

[sweet socket]

I think this should work. Sorry I’m using phone so I cannot send a code block. Just replace the messages with these ones

[clever trench]

No luck!

import openai
response = openai.ChatCompletion.create(
model='gpt-3.5-turbo',
messages=[
{'role': 'user', 'content': "Answer the question only related to the topics of health if they ask outside scope queries, say That I have been trained to answer only health related queries"},
{'role': 'user', 'content': 'I will later ask you a question about the passage I sent you before. Answer the question based on the passage, if there’s nothing relevant to the question from the passage, just reply I don’t know'},
{'role': 'user', 'content': 'python code for factorial'}
],
max_tokens=100,
temperature=0,
)

print(response)

[sweet socket]

the first user content should be the passage you ask questions about

So what’s the pyson code for factorial? it should be your question there i suppose

[uneven oak]

Look into the llama_index library @clever trench