Pro tip: I read that one AI model now uses less power than a light bulb, and I can't tell if that's a win or a problem
Saw a number last month on a tech blog that stuck with me. A small language model running on a phone chip used about 4 watts, which is less than a lot of LED bulbs. Side A says great, cheap AI everywhere means more people can use it and the big data centers stop hogging power. Side B says cheap AI everywhere just means everyone runs ten of them at once, so total power use goes up anyway. I keep going back and forth. If a model costs almost nothing to run, what stops apps from stuffing one into every single feature? Curious where people land on this. Does low power per model actually save anything overall, or does it just make us use more?