# When a user triggers multiple function calls back to back, it seems like the new model "gpt-3.5-turbo-0613" randomly decides to send imcomplete json data in the "arguments" field of the response. How can I prevent this?