Unexpected answers in API

I am running a complex ideation query through the API. It was working earlier this week, and now returns nonsense. Is this to do with the recent outage and the fix that was implemented? CHTP 3.5 turbo seems unable to remember what it was doing.

Example Prompt
I want to generate ideas for the following challenge - how might we create unique experiences for wedding parties? Generate a list of 5 famous fictional or real characters from movies or books. Provide a short description of the unique characteristics of each character. Impersonate each character, and have them generate a few highly diverse and creative ideas in response to our core challenge. Organise your response by idea including the following parameters - a short title, a short explanation connecting the fictional characters response and the idea, a more detailed explanation of how the concept works, and the key benefits to the target market. Make sure to provide details for how the concept might work, including any key features.

Answers provided through the chat interface are all wedding related and highly relevant to the prompt. Answers provided through the API are random business ideas with no relevance to the core challenge presented in the prompt.

What is going on?

try gpt 16k 0613

1 Like

If you were using gpt-3.5-turbo-0301 for its long-term reliability promise — well, OpenAI broke it in the last week. What model are you calling?

yep. are they unbreaking it any time soon?

note: i have only been specifying gpt-3.5-turbo … but I guess this still applies to me. I wonder what happened exactly

Just tried this through my Bubble app … not working though. I’ll take a look at the Bubble forums and see if there is a problem their side.

this is embarrassing … I was passing the wrong parameters into the API call … doh!

On the plus side, I have learned a bit more about available GPT models and temperature settings

1 Like