I would love to reach a point where competent language models become commodities that anyone can run on low key hardware. Having one at your disposal can open up for some gorgeous applications and workflows by the community. As it stands at present though, there are insurmountable moats or very expensive ones.