Fine-tuned ChatGPT outperforms GPT-4 in news article summarization using only synthetic data, as demonstrated by a human-level automated evaluation system. The study employed chain of density (CoD) prompting to generate synthetic data for fine-tuning purposes. While the performance of fine-tuned ChatGPT is slightly below that of GPT-4 with CoD prompting, it significantly surpasses zero-shot GPT-4 in terms of cost and latency, making it a viable option for real-world deployment. The use of synthetic data and automated evaluation systems like ScoreStringEvalChain and PairwiseStringEvalChain can enhance the capabilities of language models while maintaining performance at scale.