Water use comes out to about 150.000 chatGPT queries per quarter pounder. Using 10ml per prompt and 15.000l per kg of beef.
Still off by many orders of magnitude.
Also that’s just the running costs. If we go into training we’re looking at a comparison the other way around. Training GPT-3 cost Microsoft 700.000 liters of water. So that’s 466.6 quarter pounders.
The CO2 calculation is also useful and interesting, but they did say “water consumption”
Fair.
Water use comes out to about 150.000 chatGPT queries per quarter pounder. Using 10ml per prompt and 15.000l per kg of beef.
Still off by many orders of magnitude.
Also that’s just the running costs. If we go into training we’re looking at a comparison the other way around. Training GPT-3 cost Microsoft 700.000 liters of water. So that’s 466.6 quarter pounders.