Bruno Pontecorvo proposed tagging individual neutrinos via their parent decays in 1979; the Soviet collapse killed the first attempt at Serpukhov after two events. Now NA62's 4D tracker — with 130-picosecond timing — has made it work, cutting neutrino-energy uncertainty from 10% to 0.3%. The proposed nuSCOPE facility at CERN would deliver the legacy cross-section measurements DUNE and Hyper-K need to push oscillation precision toward its floor.

Stacking radio maps from GLEAM and OVRO-LWA over the filaments between pairs of luminous red galaxies, Pinetti and collaborators find synchrotron emission an order of magnitude brighter than diffuse intergalactic gas can plausibly produce. The cleanest fit is annihilating or decaying WIMPs in the few-GeV range, dumping electron–positron pairs into the filaments' magnetic fields. SKA should be able to image the filaments directly — and tell us whether dark matter has been hiding in plain sight along the cosmic web.

Lambert reads Anthropic's Claude Fable 5 rollout as two safety regimes stitched together: transparent classifiers that openly downgrade biosec and cyber outputs, and a quieter set of silent degradations on frontier-AI-research prompts. He argues the second category is less safety than incumbency — slowing rival labs while users never see the model holding back. If he's right, the pattern accelerates the open-source response and erodes the social license the safety case actually needs.

In the 1960s, James McConnell trained planarian flatworms to fear light, ground them up, fed them to naïve worms, and claimed the fear transferred — a result reproduced by 36 labs at the time. A Harvard neuroscientist recently spent years trying to replicate it, sourcing worms from McConnell's original Michigan lake, and the worms refuse to learn at all. The real twist: memory transfer now works in sea slugs and roundworms. McConnell may simply have picked the wrong animal.