Open-source language models and chatbots are proliferating at an incredible rate. It's still hard to train full language models, but still very good performance by instruction tuning. I'd love to better understand the comparative advantage as between these different base models. But yeah, interesting economic issue.