Wondering about the idea space: just like the article explains that there could be idea progress in physics theory, even if it’s disincentivised by the research system, there could in principle be progress in AI model space. I tend to agree with the discussion on experimental physics and GPUs
Very much worth reading: timdettmers.com/2025/12/10/w... The (unexpected) fundamental physics part seems overall accurate to me.
Lesenswert!
I feel like this is a very strong counter to a lot of arguments about fundamental economic or structural obstacles to decarbonization, or that it requires a comprehensive reorganization of our system of production. It's a political choice, that's all.
A fairly unique feature here is that imposing structure allows the algorithm to generalise to phase spaces (physics processes) it hasn’t seen in the training!
A fairly unique feature here is that imposing structure allows the algorithm to generalise to phase spaces (physics processes) it hasn’t seen in the training!
Graph based on the data in https://en.wikipedia.org/wiki/Men%27s_pole_vault_world_record_progression , showing the world record height evolution from 1965 (about 5.3m) to 2025 (6.30m). The world records by Sergey Bubka are indicated in blue. In the first part of his career, there was a rapid succession of world records and then it leveled off (second derivative of the curve is negative). After about 20 years without world records (flat line in the curve, first derivative ~0), Armand Duplantis started breaking world records, cm by cm (indicated in orange). The second derivative is positive, which means the curve bends upwards.
I naively thought: let's roughly extrapolate the height at which Mondo Duplantis's pole vault world record series would end. The second derivative turns out to be positive! 😱
#math in the wild #tokyo2025 #worldathletics
"It’s to the point that we’re well past dot com boom levels of investment, and, as Kedrosky points out, approaching railroad-levels of investment, last seen in the days of the robber barons."
INVESTIERT DOCH MAL SO IN DIE POST-FOSSILE TRANSFORMATION IHR DACKEL
👀Who needs eyes when you have DeepMET? 👉Learn how this smart algorithm developed by CMS is helping us see escaping invisible particles better than ever before:
cms.cern/news/teachin...
#machinelearning #cern @cern.bsky.social #physics
Was die @iea.org in ihrem jüngsten Bericht über Deutschland aufgeschrieben hat, ist nicht nur ein ziemlich gutes Zeugnis für die Energiepolitik der Ampel-Koalition (mit Ausnahme des Verkehrssektors) – sondern auch die klare Empfehlung an die nächste Regierung, am bisherigen Kurs festzuhalten:
Fixed it for #academia
Und wenn letzteres, warum wird diese nicht bestraft? Liegt das an den vorherigen Kompetenzzuweisungen?
Ist es nicht kurios, dass genau die als wirtschafts- und finanzpolitisch kompetent geltenden Parteien (Union, FDP) keine Gegenfinanzierung haben? Ist das Überkompensation auf der linken Seite (gelten nicht als solide, darum muss das Programm so sein) oder große Unehrlichkeit auf der anderen Seite?
Es gibt deutschsprachige Experten! bsky.app/profile/joac...
This makes me optimistic, unless one thinks that finishing and conceding high-quality chances are skill issues (which they may well be) @effzeh.bsky.social
Now only the results need to come! And some work on opponent shot locations...
Team radar 1. FC cologne (this season 2. Bundesliga)
