| ◈ Science | A 4σ Crack in the Standard Model, Written by the Physicists Who Found It | 9 min |
| ◈ Science | What Physical 'Life Force' Turns Biology's Wheels? | 14 min |
| ⬡ AI | DeepSeek V4: Almost on the Frontier, a Fraction of the Price | 7 min |
| ◉ Wildcard | Classical Statues Were Not Painted Horribly | 18 min |
LHCb's angular analysis of B → K π μμ — a flavor-changing penguin transition from a beauty quark to a strange quark — sits four standard deviations from the Standard Model, with hadronic 'charming penguin' contributions unable to absorb the gap. Two of the analysis authors walk through why this is the cleanest version of the b → sℓℓ anomaly story to date, and how the dataset already in hand (3× the analyzed sample) could push it toward 5σ. Leptoquarks remain the headline candidate.
After fifty years of partial pictures, the bacterial flagellar motor finally has a mechanism: the proton motive force, that sub-cellular electrochemical gradient, gets converted into kinetic rotation at hundreds of revolutions per second. Wolchover argues this is the closest biology comes to a Carnot-style unifying principle — entropic flow doing useful work, written in protein.
V4-Pro and V4-Flash land near frontier capability at a small fraction of frontier price — V4-Flash undercuts even OpenAI's cheapest tier at $0.14 per million input tokens. The number worth sitting with: V4-Pro reaches that capability level using roughly 27% of the per-token FLOPs of V3.2. The closed-vs-open price gap is narrowing faster than the capability gap.
The standard line on the museum-reconstruction polychrome statues that look garish to modern eyes is that ancient taste was simply different from ours. Weir's deflating counter: the reconstructions are just painted badly. Surviving ancient frescoes, mosaics, and vase work show a subtlety the reconstructions don't bother to reproduce — and the chromophobia we've inherited may be a 500-year accident of which paint happened to survive Renaissance excavation.
The European Strategy Group has formally recommended the electron–positron Future Circular Collider — a 90.7 km Higgs factory — as CERN's next flagship, with a descoped backup that trims construction cost ~15% while keeping the physics competitive. The CERN Council convenes in Budapest this month for the Strategy update vote; a construction go/no-go is expected around 2028. The decision will set the direction of the field for the next two decades.
Anthropic's Managed Agents team on the architectural call they made: separate the inference engine and harness ('the brain') from execution environments and tools ('the hands') behind durable interfaces, so neither side has to know what the other can do. The OS analogy is explicit — abstractions outlast the hardware they hid. Concrete payoff: 60–90% lower time-to-first-token, no pet containers, and credentials never crossing into the sandbox.