"ChatGPT consumes a lot of energy in the process, up to 25 times more than a Google search. Additionally, a lot of water is also used in cooling for the servers that run all that software. Per conversation of about 20 to 50 queries, half a litre of water evaporates – a small bottle, in other words."
AI is predicted to consume twice as much energy as the whole of France by 2030
Training GPT3, took 1,287 MWh (Megawatt hours) of electricity.
@gerrymcgovern Everyone's busy learning the wrong lesson from the logarithmic scaling of LLM performance. The correct lesson isn't "we need massive datacenters to gain a minor edge", the correct lesson is "a tiny little LLM that can run on a cellphone doesn't sacrifice much quality compared to a datacenter-spanning behemoth".
@carnildo
True. But it’s not about quality for the end user. It’s about control and access to input data for the vendor.
@gerrymcgovern
