#wrong answer to hi text in the api.

11 messages · Page 1 of 1 (latest)

crimson wolf
ionic sluice
#

I had noticed that too, especially with text-davnci-002, it seems to give an answer to someone else's prompt, eg "I am sorry to hear that you are having difficulty with your computer." OR "The answer to your question is that it is possible to get a restraining order against someone for texting you. In order to do so, you would need to show that the texting is causing you some type of harm, such as causing you to feel scared or harassed." in response to "hello". This is fixed in chatGPT but not the API.

hasty kelp
#

If you send only "hi" then it will auto complete it as if you typed "hi" in the playground.

You need to set up a conversation format in this case.

E.g.

If you select Bot template in the playground , the prompt is:

The following is a conversation with an AI assistant. The assistant is helpful, creative, clever, and very friendly.

Human: Hello, who are you?
AI: I am an AI created by OpenAI. How can I help you today?
Human: {USER TEXT GOES HERE}
AI:
vast isle
#

@crimson wolf

#

It's actually a completion engine.

#

So that means this would give awkward answers to you.

crimson wolf
vast isle
#

Like completes the conversation.

#

You say Hi.

#

It returns this:

#

Hi I have some problems with that software.