Fine-Tuning Chatgpt 3.5 With Synthetic Data From Gpt-4 | Very Interesting Results