GPT 3.5 refuses to answer quite often.
That’s why I implemented the retry and delete buttons. It’s better not to have failed messages in the chat history.
Dolphin will try to answer whatever you say.
You will get fewer refusals with the Dophin, but sometimes it is less intelligent than GPT 3.5 and it’s slow.
Dolphin delay issue
When you’re using the Dolphin AI, as the chat history gets longer, the response gets slower.
You can adjust this delay by limiting the context length.
The context length limit is set to 8K by default. If you want the AI to remember more, increase that value. But the response will get slower as the chat history grows.
Dolphin AI repeating sentence issue
One problem with Dolphin is that it tends to repeat the same sentences.
I think it is related to the intelligent capacity of the AI model. It looks like it repeats the same words when it has nothing more to say.
I’ve implemented a method to cut off when the program detects a repeating sentence. Because of that cut, sometimes the dialog may end abruptly.