Dolphin 3.0 !

I guess so, I used to be excited for the original dolphin's and other popular finetunes at the time, but I kind of stopped caring after a while. I think part of that is because if we look at these general finetunes where the goal is to achieve higher scores or better output in general, I feel like it's a bit of a lost cause. It's also really hard for me to take it serious when a single person claims to achieve a 5-10 point increase in benchmarks that somehow a billion(s) dollar company couldn't do. In the beginning there was *some* hope that maybe by training it on some better OpenAI data that we could improve it. And it might have worked a bit to some extend, but at this point these companies have invested so much money that if a single person who rents a few A100's could make that much of a difference, they would have done that by themselves. Think about how big Meta's datacenter is for example. They could literally run the same experiment in 2 minutes over and over and over. Not just that, but, they have a lot more advanced techniques then "just show it some different data" at this point. As far as uncensored models go, we now also have uncensored models in pretty much every niche and then in every niche of those niches. Not to mention, generally uncensored models in most areas (finetuned on external data) and then also abliterated models for those who prefer that. So personally, I'm just not interested in these anymore, but to each their own.