Generalized results in Davinci-003

API
1

Hi

My AI article tool started generating too much random content. Like When asked to write a fictional novel, instead it generates content for ‘How to’ write one. It could have been a case of wrong prompt, but earlier with similar prompts I was getting better results. Any suggestion or a way around would be highly appreciated.

Example of Generalized result-

  1. Write a fictional romantic novel. Love is a powerful emotion. It can… | by Atul Singh | Medium
  2. Fictional romantic novel. Fictional romantic novels often spark… | by Atul Singh | Medium

Example of great result by Davinci-003-

Can someone also suggest optimum Temperature, Top P, Tokens , Frequency penalty, and presence penalty for this?

2

Can someone also suggest optimum Temperature, Top P, Tokens , Frequency penalty, and presence penalty for this?

I think broadly these are the kind of things that need to be tuned on an example by example basis, it is hard to give specific advice other than try a few different options and see what works.

3

Also, worth noting, the response you get from the API should not have changed recently. Are you comparing the API results vs ChatGPT for example?

4

Hi, Thanks for the reply. No, actually I was referring to the results generated by the Davinci-003 API. Something seems off to me since a few weeks. Anyways thanks for confirming. Seems will have to tune the Temperature a bit.

1 Like