I noticed a difference between enabling knowledge on the node and using the AI task. I guess it's maybe because of the deterioration in gpt performance. When enabling knowledge on the node, which model is used ? is it gpt3.5 by default ?