The problem in finding this number, as we describe in our piece published in May, was that the AI companies are the only ones who have. We surrounded Google, Open and Microsoft, but each company refused to provide our data. Researchers we talked about the effects of AI on the energy grid compare to the comparison of a car trying to measure the fuel performance of it, enabled it to run, evaluating the engine size rumors, and it looks like the highway.
But then in this summer, after we were published, a strange thing started to happen. In June, Openi’s Sam Altman Is written The average chat of GPT uses 0.34 watts of energy for GPT queries. In July, the French AI Startup Mistles did not publish a number directly but issued a Estimate Of the emissions. In August, Google revealed that Gemini uses about 0.24 watts of energy to answer a question. Google and Openi’s data were similar to that and I estimated for medium -sized AI models.
So with this new transparency, is our job complete? Did we finally harp our white wheel, and if so, what happens to people who study the effects of AI’s climate? I reached some of my old sources and some new resources to find out.
Numbers are vague and just chat
The first thing they told me is that there is a lot of missing from the tech companies published this summer.
For example, the open number of the openings did not appear in a detailed technical dissertation, but did not appear in a blog post in the reversal that contains many non -response questions, such as what model he was referring to, how the energy use was measured, and how different it is. Google data, as Crown Heart has stated, has the average amount of energy in each question, which does not give us a sense of maximum energy -related gymnastic reaction, such as when it uses a reasoning model for “thinking” through a difficult problem or produces a really long response.
The numbers only refer to interaction with chat boats, not in other ways that people are relying on Generative AI.
“As video and icon are used by more prominent and more people, we need a number of different ways and how they measure the AI platform,” says AI and climate lead Sasha Losoni on the AI platform.