|
|
|
|
|
|
by williamcotton
1114 days ago
|
|
|
It’s doing so because it includes the entire history of choices in every prompt. By using the API directly and modifying the response to only include the assistant messages and not the user messages it would stop this behavior. |
|
|