The big deal here is releasing the model pre-trained. The training part would cost $1M in compute costs if you had to do it in-house. The idea is that a compan

The big deal here is releasing the model pre-trained. The training part would cost $1M in compute costs if you had to do it in-house. The idea is that a company or developer downloads the model already trained on natural language. Then you add your additional documents and data and prompt it for answers internally, rather than through a service like OpenAI.