Mistral fine-tunes are competitive with GPT 3.5 and 4 on many, many tasks, people are just enamored with the (admittedly impressive) ability of very large gener

Mistral fine-tunes are competitive with GPT 3.5 and 4 on many, many tasks, people are just enamored with the (admittedly impressive) ability of very large general models. In terms of business applications, the range of things that need GPT-4 to solve vs. Mistral 7B is diminishing small and infinitesimal when you factor in Mixtral. I’m all for doing advanced research to create the most interesting and powerful thing possible, but in more practical terms, OpenAI has zero moat. Once the infrastructure around open source stabilizes and we get a few more rounds of improvement to efficiency, I think this will be plainly obvious.