Anthropic rounds $900B, quantum routing solves chip congestion, time finally works in JavaScript

Anthropic is moving fast. The AI company is asking investors to commit capital for a potential $900B+ valuation round within 48 hours, sources say. If it closes within two weeks as expected, it marks another swing in the venture funding cycle-from caution six months ago to aggressive deployment of capital into the handful of companies perceived as frontier. The company's latest fundraise would value it higher than most Fortune 500 firms, despite having shipped no consumer product and limited public information about unit economics.

When Your Clock Breaks Your Code

Date and time handling in JavaScript has been broken for years. There's no proper way to handle timezones, daylight saving time, or even simple arithmetic without pulling in a library. A new Stack Overflow conversation with Jason Williams, creator of the Rust-based JavaScript engine Boa, digs into why this matters and how the Temporal proposal-finally moving toward standardisation-could fix it. For developers building payment systems, scheduling tools, or anything that touches timezones, this is the relief they've waited for. Temporal lets you work with dates as first-class objects instead of the fragile string conversions that cause silent bugs across distributed systems.

Quantum Chips Get Faster Routing

Quantum chip architecture has a congestion problem. Moving quantum gates across qubits requires expensive SWAP operations that pile up overhead. A new paper from arXiv proposes swap-free routing using spectral qudits-higher levels of quantum states as orthogonal buses that carry control information without moving the computational state. The result: nonlocal operations over a path of length L drop from 3L routing primitives to 2L+1, and multiple controls arriving on distinct buses can trigger Boolean fan-in without crosstalk. It's the kind of architectural breakthrough that matters less for current 100-qubit systems and more for the thousand-qubit machines that might actually solve real problems.

Sleep data is becoming a clinical tool. Beacon Biosignals, founded by MIT researchers Jake Donoghue and Jarrett Revels, is using lightweight EEG headbands to monitor brain activity during home sleep and feed the data into machine learning systems that detect disease progression, screen for neurodegeneration, and identify patient cohorts for trials. The company's FDA-cleared device has already appeared in 40+ clinical trials for conditions ranging from schizophrenia to Parkinson's. The insight: neural activity during sleep is more structured than waking activity, making it a high-signal window for understanding how drugs affect the brain and catching early signs of Alzheimer's or Parkinson's years before symptoms show. With a $97M raise last year and acquisitions expanding home testing access, this is infrastructure shifting from lab to routine.

For teams shipping AI features behind opt-in toggles-"Try our agent mode," "Enable code suggestions"-a new tutorial walks through propensity score methods that separate adoption bias from the feature's actual effect. When power users opt in at 65% and light users at 12%, a raw comparison inflates the true +8 percentage-point effect to +21pp. The fix: reweight or rematch your comparison groups using propensity scores so opted-in and non-opted-in users look comparable on observables, mimicking what a randomised experiment would have given you. The full tutorial includes code, balance diagnostics, and bootstrap confidence intervals-practical tools for anyone running causal inference on LLM features in production.