Gpt-3.5-turbo-0613 answers question beyond restriction

In 3.5-0301, I can restrict the assistant answer in a specific area by making constraint in system message. But for 0613, the answer could easily go beyond the restriction and start to answer any questions.
It becomes less steerable in this sense…

It seems like even with the update, the system message is not a strong command. Something about it gets blurred out, and it’s only a soft suggestion.

I am getting the error “maximum context length is 4097 tokens” even if i am using the 0613. It wasn’t like that couple of days ago…

Use gpt-3.5-turbo-16k or gpt-3.5-turbo-16k-0613 instead.

1 Like