Open-source models like Mixtral beat proprietary AI on two fronts: you can run them locally for privacy and control, and fine-tune them cheaply with LoRA. That's where the real economic value is.
What's open source AI good for as compared with the big folks like OpenAI or Google or Cohere, Anthropic, et cetera? What about using these LLaMA models from Mixtral or some of these others? The big advantage of the open source models to my mind is one, being able to run them more locally, being able to have control over where they're going to happen. This allows for greater privacy and control and also integration into our everyday lives. That could really be a level up. The second thing I think is important about them is the idea of fine-tunability, being able to customize the way that we use them and what is actually even inside them. This is where LoRA fine-tuning and using Hugging Face AutoTrain become super, super powerful in the current context. Taking advantage of what they do well that the others just can't do is going to be the secret to getting good economic value out of these open-source elements.
