So my GPU is about 300 watts and a still blatantly stupid LLM can write a little faster than me. Take off 100w to bring that down to my own writing speed then make it 10x slower to turn that 200 watts into 20 watts. Even with that heavy bias in the LLMs favour (forgiving it the entire power cost of my PCs other components that it partially utilizes) what we get is something slow, dumb, and incapable of learning because any local model is statically weighted.
Now, take your privileged writer status human brain and factor in all the other power required to keep it comfy in an air conditioned room, the labor required to put a roof over your head, keep your home plumbing working, make your food, deliver you pen and paper to write with - or are you using an electrically powered appliance to record and later communicate your thoughts? Oh, did you need to go to sleep for a while?
So my GPU is about 300 watts and a still blatantly stupid LLM can write a little faster than me. Take off 100w to bring that down to my own writing speed then make it 10x slower to turn that 200 watts into 20 watts. Even with that heavy bias in the LLMs favour (forgiving it the entire power cost of my PCs other components that it partially utilizes) what we get is something slow, dumb, and incapable of learning because any local model is statically weighted.
That’s one way to compare it.
Now, take your privileged writer status human brain and factor in all the other power required to keep it comfy in an air conditioned room, the labor required to put a roof over your head, keep your home plumbing working, make your food, deliver you pen and paper to write with - or are you using an electrically powered appliance to record and later communicate your thoughts? Oh, did you need to go to sleep for a while?
Do I get to include the gargantuan cooling system for the data centre and all the infastructure required to keep that going?
Total impact is the only fair comparison. Pollution from the power plants included, sewage treatment from the houses too.