> People are often curious about how much energy a ChatGPT query uses; the average query uses about 0.34 watt-hours
My point was twofold, “the average query” becomes less meaningful as the variance increases. Sure one can _in principle_ report W-h spent on your account or query, but I think it will get more opaque and hard to predict. Average becomes less useful when agents can do an unpredictable amount of work with increasingly large bounds.
Second, and this was perhaps worthy of expanding in my post - I think in the radical abundance world that Altman is describing, energy and efficiency stops being something that people talk about. Fusion, space based solar farms, whatever, it’s easy to imagine solving this stuff. And I think you can even imagine this happening sooner if for example one provider stamps a “100% renewable datacenter” option on their AI product. Then you might not care about energy efficiency at all; in which case you just care about profit.
My point was twofold, “the average query” becomes less meaningful as the variance increases. Sure one can _in principle_ report W-h spent on your account or query, but I think it will get more opaque and hard to predict. Average becomes less useful when agents can do an unpredictable amount of work with increasingly large bounds.
Second, and this was perhaps worthy of expanding in my post - I think in the radical abundance world that Altman is describing, energy and efficiency stops being something that people talk about. Fusion, space based solar farms, whatever, it’s easy to imagine solving this stuff. And I think you can even imagine this happening sooner if for example one provider stamps a “100% renewable datacenter” option on their AI product. Then you might not care about energy efficiency at all; in which case you just care about profit.