A Magnetic Mystery, Vanishing Clouds, & the Cost of Reason
This week’s science bits from SWTG
Scientists Stumped by Link Between Magnetic Field and Oxygen Levels
I don’t find this terribly convincing, but who am I to judge? VGADM stands for Virtual Geomagnetic Axial Dipole Moment. Figure: Kuang et al, Sci Adv 11, 24 (2025).
In a study that just appeared in Science Advances, geoscientists report a mysterious correlation between the strength of Earth’s magnetic field and atmospheric oxygen levels in the past 540 million years. They speculate that tectonic dynamics might both have influenced Earth’s magnetic field as well as the abundance of oxygen-binding organisms or minerals, but really they have no idea what’s going on. Maybe it’s worth pausing work on your theory of everything for a moment to take this on. Paper here, more here.
More Evidence That Earth Is Losing Clouds
Image: NASA Worldview
Clouds are the Achilles’ heel of climate models. Their formation and behavior is beyond the models’ capability to compute, and yet their response to warming temperatures is a major feedback mechanism. A new analysis of satellite data from NASA just confirmed that the world is losing clouds which reflect sunlight back into space. The analysis shows that cloud regions have contracted especially around storm zones. This is bad news, as it makes it yet again likelier that we are underestimating the pace of warming. Paper here, press release here.
Reasoning Models Take Up a Lot of Energy
Figure: Dauner & Socher, Front. Commun. 10 (2025)
In a finding that surprises absolutely no one, researchers from Germany confirmed that Large Language Models consume considerably more energy, and produce considerably more carbon emissions, if they use extra reasoning steps. The most carbon-intensive of the models they tested (which does not include either GPT, Gemini, or Claude) is DeepSeek, which on their questions, however, also reached the highest accuracy. The researchers want you to carefully consider whether demanding more accuracy from an AI is really worth the extra carbon footprint, but I don’t have the energy to think about it… Paper here, press release here.