Anyone know of any data comparing the energy and water usage of a human being performing a cognitive task - like reading and summarising an article - to genAI doing the same thing?
I suspect the latter would be guesses because these companies are deliberately trying to hide the true costs. But my guess is we will find at least an order of magnitude difference.
@emilymbender Thanks! I will have a look but not sure why ill-formed?
I was quite deliberately avoiding issues around training and only asking about energy use of models once released, if that is what you had in mind. I.e. suppose we accept the direct comparability of a human performing a cognitive task and an LLM doing the 'same thing', which is what all economic predictions seem to be based on, then which is more energy and water efficient?
@tomstoneham We get into it in the podcast episode I linked to but basically: humans exist (and have a right to do so) and by existing we consume energy. So comparisons between humans (existing and) doing some task and LLMs doing the task are spurious.
@tomstoneham Unless you can find a way to calculate only the additional energy required by a person to do a task, the comparison is not just spurious, but drastically dehumanizing. A person's "energy and water usage" is a condition of their existence, not the cost of their doing some task.
Add comment