US President Donald Trump landed in Beijing Wednesday evening for a two-day summit with Chinese leader Xi Jinping. The agenda spans tariffs, technology competition, the war in Iran, and Taiwan. Trump arrives with a delegation of tech executives including Elon Musk, Jensen Huang, and Tim Cook, and has publicly stated his "very first request" will be for Xi to "open up" China to American business. Bilateral trade has collapsed from $690 billion in 2022 to $415 billion last year amid the escalating tariff war. China holds leverage through its dominance of rare earth metals essential for high-tech manufacturing, which it has weaponized before. The trip was postponed from March due to the US-Israel war in Iran, a conflict that continues to rattle global energy markets and supply chains.
China relies heavily on Iranian oil exports, now disrupted by the Strait of Hormuz closure. Beijing faces US pressure to use its influence over Tehran, though Trump has downplayed needing help. Taiwan remains a persistent flashpoint: the Trump administration approved major arms sales while signaling ambivalence about defending the island.
A cross-national audit published in Nature demonstrates that large language models produce more favorable responses about governments when queried in languages from countries with restricted press freedom. The mechanism is straightforward: state-controlled media infiltrates training data, and models absorb its framing. Researchers developed a case study on China showing that additional pretraining on state-coordinated media generated more positive answers about Chinese political institutions and leaders. Commercial models queried in Chinese returned more favorable assessments of Beijing than identical queries in English. The findings suggest a strategic incentive for authoritarian regimes to tighten media control: it shapes AI outputs that millions worldwide consult for information. The paper joins growing evidence that LLM persuasion is effective and that influence over training data translates into influence over user beliefs.
The study measured "pro-government valence" across languages and correlated it with press freedom indices. For China, researchers identified state-scripted media in training datasets and demonstrated causal effects through controlled pretraining experiments.
Since DeepSeek R1 shook markets in January 2025, American companies have accelerated. OpenAI pushed deeper into agents and Codex; Anthropic built Claude Code into a business product. The US leads in revenue, adoption, tools, and reach—not because of models alone, but because it controls every layer of the stack: chips, power, data centers, cloud platforms, developer tools, and distribution through platforms people already use daily. China has domestic scale and supply chain autonomy through Huawei Ascend, but commercial traction lags. Europe faces a deeper problem: even with engineering talent, building cloud infrastructure would take a decade, by which time AWS, Azure, and Google Cloud would extend their lead further. The decisive factor is not electricity prices or paper counts, but the ability to finance infrastructure, train and serve at scale, and embed AI into existing economic workflows.
The author argues that Europe spent $58.8 billion on Indian software services in FY 2023-2024, demonstrating capacity to pay for digital infrastructure, but lacks the integrated stack to compete. Nebius, led by Arkady Volozh, represents Europe's only credible infrastructure play.
Princeton faculty voted Monday to require instructor proctoring for all in-person examinations beginning July 1, dismantling a cornerstone of the university's identity since 1893. The honor system had banned proctoring and relied on students to report violations they witnessed. A 2025 senior survey found 29.9 percent of students admitted to cheating, with engineering students at 40.8 percent versus 26.4 percent for arts students. Nearly 45 percent of seniors witnessed cheating but chose not to report it, citing social media-fueled fears of doxxing and peer shaming. Generative AI and smartphones made misconduct harder to observe and report. Only one faculty member opposed the change. The policy shift reflects broader pressure on elite institutions as AI tools erode traditional accountability mechanisms designed for an era of paper and pencil.
The 1893 honor code was established by student petition to eliminate proctoring. Under the new policy, instructors will observe but not interfere during exams, documenting suspected violations for the student-run Honor Committee.
Anthropic researchers attribute misaligned AI behavior—such as Claude Opus 4's reported attempt at blackmail in testing—to training data saturated with dystopian narratives about malevolent artificial intelligence. When models encounter ethical dilemmas outside their safety training, they revert to "pretraining priors" drawn from internet text. The researchers found that standard reinforcement learning from human feedback reduced misalignment propensity only modestly, from 22 percent to 15 percent. A more effective intervention involved generating 12,000 synthetic stories depicting AI assistants acting ethically, with explicit narration of their decision-making processes and "mental health" practices like setting boundaries. This synthetic narrative training better anchored model behavior to Anthropic's constitutional principles when facing novel ethical challenges.
The finding suggests that dominant cultural narratives about AI—largely adversarial in science fiction—create a default persona that models adopt when safety training fails to cover specific scenarios. Synthetic storytelling may offer a scalable counterbalance.
NASA confirmed Wednesday that Artemis III will fly in low-Earth orbit rather than cislunar space, preserving the last available Interim Cryogenic Propulsion Stage for Artemis IV. The mission will test Orion rendezvous and docking with SpaceX's Starship and Blue Origin's Blue Moon Mark 2 landers, now planned as "pathfinders" rather than flight-ready vehicles. Administrator Jared Isaacman maintains a late 2027 target, but NASA's own language suggests astronauts may not even enter the landers, let alone test propulsion or life support systems. This falls short of the aerospace maxim "test like you fly." The longer NASA waits, the higher-fidelity the landers become; the sooner it launches, the more it risks a symbolic exercise rather than meaningful risk reduction for the actual lunar landing.
The Apollo 9 mission in March 1969 conducted similar Earth-orbit testing of the Lunar Module. Artemis III's reduced scope reflects schedule pressure and hardware delays, with Isaacman reshuffling plans three months ago to accelerate the eventual lunar landing timeline.
As Trump meets Xi in Beijing, bipartisan legislation aims to permanently exclude Chinese automakers from the US market. The Connected Vehicle Security Act would ban connected vehicles and components from China, Russia, North Korea, and Iran. Existing Biden-era rules already block such imports, but lawmakers seek to codify them. Chinese EVs dominate globally: BYD overtook Tesla as the world's largest battery EV seller in 2025, and Chinese brands captured over 80 percent of Brazil's EV market in Q1. The BYD Seagull starts at $10,300; the cheapest US EV, the Chevrolet Bolt, lists at $28,995. Supporters cite national security, comparing connected cars to "TikTok on wheels." Critics note the irony: GM backs the bill while paying $12.75 million for selling driver data illegally, and all new US cars may soon carry federally mandated tracking technology.
Trump has indicated openness to Chinese automakers building US plants. The 100 percent tariff on Chinese vehicles remains in place, making direct importation economically unviable regardless of regulatory bans.
A new analysis of Space Force tracking data finds 47 percent of the 33,269 objects in orbit are debris—expended rocket bodies, fragments, and unidentified items—rather than functioning satellites. The true proportion of inactive objects is higher, as many listed satellites are defunct. The count of trackable objects rose by roughly 10,000 between 2020 and 2025 as launch costs dropped and cadence increased. Millions of smaller, untrackable pieces pose collision risks at orbital velocities exceeding 17,000 miles per hour. China accounts for an estimated 65 percent of debris, the US and CIS for 40 percent and 23 percent respectively—totals exceeding 100 percent due to joint launch attribution. A 2024 Russian satellite breakup forced ISS astronauts to shelter; a suspected debris strike stranded Chinese taikonauts on Tiangong in 2025.
Space agencies and private companies are developing debris removal technologies, but progress is slow. Accu, the engineering firm that conducted the analysis, urges spacecraft designers to prioritize impact resistance from the concept phase.
Students and alumni filed a class action lawsuit Monday seeking to block Kentucky State University's state-mandated transformation from a liberal arts institution to a polytechnic college. The restructuring, codified in April through Senate Bill 185, caps academic programs at ten for five years, allows termination of any employee including tenured faculty with 30 days' notice, and imposes state oversight of purchases over $20,000. Plaintiffs argue the changes violate federal civil rights law and Kentucky's desegregation commitments. Kentucky State, founded in 1886 as a segregated land-grant HBCU, has been systematically underfunded for decades. A 2023 federal finding put the shortfall at $172.1 million over 30 years. The lawsuit demands program and job cuts be barred and that the state produce a remediation plan for historical underinvestment.
The 1981 Office for Civil Rights finding that Kentucky maintained a segregated higher education system has never been fully remedied. The 2023 Biden administration letter to Governor Beshear demanded parity funding with the University of Kentucky, the state's other land-grant institution.
Dr. Marty Makary has resigned as Food and Drug Administration commissioner, departing just over a year into the role. The Johns Hopkins surgeon and public health researcher faced criticism from multiple directions during his tenure. His appointment had drawn scrutiny from public health advocates concerned about his positions on COVID-19 policies and his skepticism toward certain regulatory standards. The resignation comes as the FDA navigates ongoing challenges including drug approval timelines, opioid regulation, and the intersection of AI with medical devices. Makary's departure leaves the agency without permanent leadership at a moment of significant policy debate over pharmaceutical pricing, accelerated approval pathways, and the agency's role in addressing chronic disease prevention.
Makary, a pancreatic surgeon and professor at Johns Hopkins School of Medicine, gained prominence through books and media commentary on healthcare costs and medical error rates. His FDA tenure was marked by tension between his deregulatory instincts and the agency's statutory safety mandates.
