“Fine-tuning open source base language models on ChatGPT outputs will likely result in models with significant factual accuracy problems.”