I used to think running your own local model is silly because it’s slow and expensive, but the nerfing of ChatGPT and Gemini is so aggressive it’s starting to make a lot more sense. I want the smartest model, and I don’t want to second guess some potentially quantized black box.