chatGPT seems to "understand" better than davinci003. Why and how to replicate chatGPT's quality?

lowjootat · 2023-02-06T05:31:58.824Z

I have a use case where I want to ask GPT to rate the helpfulness of a video (based on the transcript, description, title) to an specific audience.

I am finding that for the exact same prompt, chatGPT works much better than davinci003, with chatGPT making more accurate judgements regarding the helpfulness.

Any idea why this is so?

If I want to get the same level of accuracy that chatGPT has, any suggestions for how I can achieve that via an API? Thank you!

ruby_coder · 2023-02-06T05:34:04.233Z

lowjootat:

If I want to get the same level of accuracy that chatGPT has, any suggestions for how I can achieve that via an API?

You can wait for the ChatGPT API to be released

Coming soon… so we have been told.

oleg1 · 2023-02-07T06:33:22.439Z

Wait for Google’s response and we will have choices like never before They will not go down quietly. Maybe at all. Their “Bard” (LaMDA) is on the way and is updated every few minutes + searching the web. Few days max and we will see the response

ruby_coder · 2023-02-07T06:45:10.384Z

oleg1:

Few days max and we will see the response

Yeah, I’m using some of Google’s code to calculate embeddings (vectors), and it’s fully free, open source code.

I hope Google comes out swinging with some very competitive APIs to develop with.