Perhaps we will provide feedback to open source Llama using ChatGPT. The cost to adjust the model is presumably what's hard?