Google recently released DiffusionGemma, and it's weird in the best way.
Most of the worries I had before going local turned out to be smaller than expected. The quality of local LLMs hold up better than reviews led me to believe, and performance on 8GB of VRAM has been ...