Basically if it's been touched by tech bros with gazillion dollar valuations, it's bad. If you roll your own with some care and attention, it's good. The future belongs to those who don't outsource their thinking.

It is costly to train from scratch. Fine tuning makes more sense for me. Not all the llms are super terrible. Llama models are ranking higher than the rest but for sure they are not the optimal. Generally western models are doing better.