I ran 14b and 20b local AI models on my laptop for the same tasks, and result was surprising

Benchmark winners don't always make the best assistants.