Crazy stuff. Guys from Stanford fine-tuned Meta's LLaMa model using OpenAI's text-davinci-003 model (improved InstructGPT). The result is a powerful model that performs many tasks on par with OpenAI's model. But the cost of training was just $600. https://youtu.be/xslW5sQOkC8 https://crfm.stanford.edu/2023/03/13/alpaca.html