Should You Use Open Source or Proprietary Large Language Models?

[@mikefabrizio]

Thank you!

[@AskALJ]

How difficult is it train the open source model on premise with domain specific data such that it approaches the proprietary model in accuracy. Obviously would depend on the use case, however, would you say it is certainly doable to take an open model, train it in a specific domain and get a result (at least within that specific domain) that is significantly better than the proprietary model? Is there a yardstick benchmark that would allow a business to identify which open model may be best for various internal use cases so that business owners can make more informed decisions on which model to invest their time in to create value?

[@dwoodlock]

I would say that it is difficult enough, in initial investment and then ongoing monitoring and training, that you should first look to a foundation model + good context (RAG) + good prompt engineering to solve your problem first. That is your better and easier route to improvement.