#Policy# | 🛍️✈️ #Hengqin# Shopping Upgrade! On July 1, Hengqin launched a new departure tax refund service across 74 registered stores, including Chimelong Ocean Kingdom. Global travelers spending 200 yuan or more can use a streamlined "one bag, one code" paperless system, bypassing unpacked customs inspections. Backed by a new three-year subsidy plan (2026–2028), Hengqin is partnering with #Macao# and #Zhuhai# to build a world-class cross-border shopping paradise. Dive in: #TaxRefund# #ChinaTravel# #DutyFree# #China# #Guangdong#

“Policymakers who assume Americans will uncritically embrace AI, labor-market disruptions and all, will be in for a rude awakening,” write James Davidson and Matthew Slaughter.

POLICY: South Korea's FSC plans to release detailed tokenized securities rules in July, covering tokenized stocks, bonds and money market funds, ahead of a full regulatory framework taking effect in February 2027.

Policy signal, without the noise. Launching the new @EntEthAlliance Policy Friday tracker in beta: a live feed tracking regulatory moves from official agencies shaping institutional @ethereum . Scans, filters, stores outputs from global agencies (SEC, MAS, OCC, TREAS, ECB, CSRC...) Feedbacks welcome

Reply policy 2026: 1. Sometimes I leave my threads open here; sometimes I make them for my 2,200 subscribers (so I can read all the comments and get to know folks by name)... and sometimes I start them open, then go subscriber-only. 2. We give the $ 1-a-month subscriptions to charity or to people I think are deserving of support on this website — I don't take any of it personally. 3. 100% of the time, I block rude, spam, bot, or burner accounts. If you were a jerk off and want to be unbanned, you can always email me an apology, and I will give you a redo.

Monetary policy does not travel only through borrowing costs or the sectors that households touch first. New Dallas Fed research shows it also moves through the web of supplier relationships that ties industries together.

On-policy distillation has the same systems bottleneck as RL: rollouts dominate training time on reasoning workloads. Going async fixes throughput but feeds the learner stale-policy data, and what staleness does to OPD specifically was unstudied. The clean finding is that KL direction decides robustness. Teacher-weighted forward KL shrugs off stale rollouts, student-weighted reverse KL breaks under them, and for the reverse-KL case nothing from async RL beats just recomputing the signal under the current student. Finite teacher-score caches then turn the estimator into a bias-variance tradeoff, which is the case for multi-sample Monte Carlo. AsyncOPD: How Stale Can On-Policy Distillation Be? Paper: