On 19.08.26 10:27, Matthias Urlichs wrote:
On 18.08.26 23:21, Zlatan Todoric wrote:Source? IMHO that number is highly suspect.(some cities reported already an average increase of 4C because of AI datacenters),
I asked Claude how fast the air above a 100-MW data center rises, assuming sea level, 20 °C and no cross wind. The answer is (roughly) 5…8 m/s, or Beaufort 4, and it'd be less than 21°C warm at 500m (~1500ft).
If the data center emits a whole whooping gigawatt, we can multiply the plume's speed by two, roughly. Bft 5. The temperature is higher and ends up at ~23°C at 1km. If we then assume a stiff cross wind of 10 m/s the center of the plume would be >300m up at 1 km distance.
It is thus highly unlikely that the warm air ends up on ground level in the next-closest city, even if we assume a strong wind going in the right (or wrong …) direction. You simply can't keep that amount of hot air on the ground. Somewhat similar to a hot-air balloon.
NB: Yeah I asked an AI. So sue me, but I don't have the time, or the motivation, or the physics background, to calculate that kind of thing myself. Also, if I did it'd most likely require more power to do it than Claude Sonnet.
NB2: IMHO this kind of AI answer is rather unlikely to be diluted by cross-training. While I agree that self-reinforcing AI sloppiness is a major concern, esp. as far as textual output is concerned, that should be a fixable problem. If nothing else succeeds (Pangram comes to mind …), well, we can simply not train on prose written after 2022. Or 2026, given that right now Pangram is still reasonably reliable.
-- -- regards -- -- Matthias Urlichs
OpenPGP_signature.asc
Description: OpenPGP digital signature
