I want to know who's successfully running AI locally and getting REAL value out of it. I setup #hermes against gemma4 (lmstudio/macmini). It's taken 2 hours of back and forth for it not to get METAR weather. I feel like I'm using the original chatgpt again, and not in a good way.