I think ternary LLMs are not just a gimmick because making AI cheaper and smaller seems genuinely useful for running it on everyday devices. Maybe accuracy takes a hit, but that feels like a tradeoff worth exploring. I am not sure how to argue it better than that.