Ten years after the first chirp, the LIGO–Virgo–KAGRA network has released GWTC-5.0: 161 new events from the O4b run, bringing the running total to 390, with the sharpest sky localization yet (6 square degrees) and apparent evidence for second-generation black holes — mergers of holes that were themselves born in mergers. Siegel uses the milestone to argue that the field has crossed from heroic one-off detections into routine population science, the way optical astronomy did a century ago. A satisfying long view for anyone who remembers when a single detection was front-page news.

Last year a Yonsei University team argued that once you account for the ages of supernova host galaxies, the evidence for cosmic acceleration weakens — a result that, taken at face value, would have unseated dark energy. A new rebuttal led by Phil Wiseman, with Nobel laureates Adam Riess and Brian Schmidt as co-authors, pins down two specific errors: the analysis dropped the standard host-galaxy mass correction and mismodeled the age dependence it claimed to find. Correct both and acceleration returns fully intact — a clean case study in how the Type Ia distance ladder actually gets stress-tested.

On June 12, Anthropic's Fable 5 and Mythos 5 went dark after a Commerce Department export-control directive barred their use by any foreign national, anywhere — reportedly triggered by a jailbreak of Fable's cyber guardrails. Lambert reads the episode as a starting gun: a moment when frontier models start being governed less like software products and more like controlled munitions, with availability subject to direct state intervention. Whatever you make of the politics, the mechanism is the story.

DNA-language foundation models are being trained much the way GPT was trained on text — and Philip Ball lays out why the analogy may break down. The genome isn't a static string to be decoded; it's a physically folded, context-dependent system where the same sequence does different things depending on chromatin state, cellular signals, and 3D geometry. A useful corrective to 'just scale the sequence data' optimism, and a reminder that biology keeps some of its information in its physics, not its letters.

Persi Diaconis's famous result — seven riffle shuffles suffice to randomize a 52-card deck — quietly assumed mathematically perfect cuts. A new proof handles the realistic case of clumsy, human shuffles, and the count it lands on is pleasingly close. A clean little window into mixing times and how idealized math gets dragged back to reality.

Dattani's monthly biomedicine dispatch: an mRNA cancer vaccine that halved melanoma recurrence versus Keytruda alone in a five-year phase 2b trial, one-shot gene editing for cholesterol, and newly characterized ancestral CRISPR systems. A fast, high-signal way to keep a finger on the field without the press-release hype.