Adding Polite on the response giving underscore or blank words

I tried adding Polite Response as a stop sequence. However, The bot is giving me an underscore if there’s the bot do not know the answer. Here’s an example scenario:

Q: What is your name?
Polite Response: My name is _________.

Does anyone knows how to get rid of this but still use Polite Response?

Yeah, it’s pretty tricky to get it to not answer with an underscore. Your results above are very reproducible.

Q: What is your name?
Polite response: My name is _____.

One solution I found was to provide example questions and responses beforehand.

Q: How old are you?
Polite response: I’m 19 years old.

Q: Are you currently employed?
Polite response: I’m not employed at the moment, no.

Q: What is your name?
Polite response: My name is John

But these previous answers bias the results.

Question: How old are you?
Polite response: I just turned 52.

Question: Are you currently employed?
Polite response: Yes, I work as a secretary.

Question: What is your name?
Polite response: My name is Sarah.

I tried to simulate this effect with objective questions, but it didn’t give.

Question: What is the capital of California?
Polite response: The capital of California is Sacramento

Question: What is the scientific name for humans?
Polite response: Scientifically, humans are referred to as homo sapiens.

Question: What is your name?
Polite response: My name is ______.

I feel that the root of the problem is like you said: the model doesn’t know how to answer the question. The desired response depends on the context. If this is a customer service bot, you’ll need to find some way around this, or prevent users from asking personal questions. If your bot really does have a name, then you could include it and solve the issue that way.

Question (User): What is your name?
Polite response (BrightBot): My name is BrightBot. It’s nice to meet you!

1 Like

Thank your for your response. I appreciate it a lot. As a workaround. I just added a polite adjective to the bot introductory message so that I can get rid of the “Polite Response” as a stop sequence and just do a simple “Response”.

Thanks for your response. I appreciate it a lot.

You’re making the prompt look like a fill-the-blank test. Instead, you should have “polite response” in the instructions at the top. Also, it helps if you give the chatbot a role or objective beyond just to be polite.

3 Likes

As @daveshapautomator already pointed out, asking the bot to be polite in the instructions will do.

Adding to that, please remove the “Polite Response” from the prompt, instead use Q’s and A’s as writing “Polite Response” everytime will consume tokens and add to your bill.

Q:
A:

Here’s OpenAI API docs on how to make responses polite.

2 Likes