What completion settings does chatGPT use?

I’m getting good but not stellar results using the completion API. I notice that the results I get from ChatGPT are significantly better. Does anyone know the settings used by ChatGPT?

And, relatedly, does anyone know how they keep context going in ChatGPT? You can type “now write that in spanish” or whatever and it’ll rewrite the last response. I don’t believe the API will allow me to send a prompt long enough to include the entire last response, so I’m not clear how this is done.

Anyone?

1 Like

Uneducated guess that gpt3.5 has a higher token limit. Maybe 8k or 16k

Is gpt3.5 definitely a thing? Was not aware.

1 Like

It has been confirmed in writing by openai themselves - refer to item 2

They have also stated that chatgpt refers to the last 3000 words or 4000 tokens

1 Like

Super interesting. Thanks for this.

Would still be interested to know what sort of settings they’re using for temperature, best_of, etc.