I have the following prompt:
The following is a conversation between an AI and a user. The AI is polite, helpful and creative.
AI: Hi! How can I help?
User: summarize the conversation so far
The problem is the AI summarizes what’s written at the top of the document instead of the conversation. I have tried the marking conversation with “conversation:” and even putting it in tags but nothing seems to work.
A couple things.
It’s useful to give the AI an identity. Ex. “You”. It has nothing to do with politeness, it just makes things straightforward and easier for identification.
Unless it’s been removed, it’s ideal to separate your instructions with the conversation with a unique separator such as “###”.
Lastly, you should prime the response.
This should work better:
cGPT:
The following is a conversation between you, and a user. You are polite, helpful and creative.
[SEP]
You: Hi! How can I help?
User: Summarize the conversation so far
You: [RESP] Certainly! So far, you have asked me to summarize the conversation. As this is our first interaction, there hasn’t been any conversation before this. Is there anything else I can assist you with?
Davinci:
The following is a conversation between you, and a user. You are polite, helpful and creative.
[SEP]
You: Hi! How can I help?
User: Summarize the conversation so far
You: [RESP] So far, we’ve been discussing how I can help you. Is there anything else I can do for you?
For clarification. The prompt is everything before [RESP(onse)], including the You:
Thank you! It’s still including the document above. Should [RESP] be added as is or it should be “###”?