Yascha Mounk: “One of the strangest things about an American university education is that it bundles a whole lot of different things together. If you want to take a course in high-level mathematics or listen to the lectures of an accomplished historian of Ming China, you must also buy membership in one of the country’s most lavish gyms; purchase all-you-can-eat meals at breakfast, lunch, and dinner; rent a room in a student dorm even if your parents happen to live a few miles down the road; pay an army of administrators whose jobs range from organizing parties to hiring puppies for you to pet during finals; subsidize student clubs that are devoted to such varied activities as playing Dungeons & Dragons or blind-tasting exclusive wines; and help to pay the lavish salaries of football and waterpolo coaches…If students could attend Harvard or Rice without having to purchase the world’s most exclusive and extended all-in-vacation package, it would be much more affordable to attend these institutions.” [via Arnold Kling]
Singularity: “The next thing is still out there. There are certainly those who believe generative AI based on transformers—the architecture that kicked this all off—will still get us to vastly more powerful systems, even to vaguely defined levels like artificial general intelligence or superintelligence. But other deeply respected voices believe we’ll need at least a few more big breakthroughs. Ilya Sutskever, a godfather of deep learning and cofounder and former chief scientist at OpenAI, believes gains from scaling pretraining are plateauing as the industry runs out of data. “There’s only one internet,” he said at NeurIPS last December. Google DeepMind’s Demis Hassabis also told Wired last year we’ll need maximum scale and new discoveries. In June, he said his team has “three or four promising ideas that could mature into” the next transformer. “The 2010s were the age of scaling, now we’re back in the age of wonder and discovery once again,” Sutskever told Reuters. “Everyone is looking for the next thing.””
WSJ: “These days, Hangzhou is an AI hub at the center of China’s global tech ambitions. Its breakthrough moment came earlier this year, when local company DeepSeek shocked the world with an AI model that rivaled American programs, at a much lower cost. “All of Hangzhou went into a frenzy,” said Zhao Ji, a 41-year-old entrepreneur who is building an AI startup from his apartment in the city. DeepSeek’s shot to fame was no fluke. The capital of the eastern Zhejiang province and a prosperous city of roughly 13 million people, Hangzhou has spent decades cultivating entrepreneurship. Supportive government policies, research universities, major tech companies such as Alibaba and NetEase and relatively low living costs compared with Beijing and Shanghai have made Hangzhou a mecca for tech talent.”
Christoph Schweizer (BCG newsletter): “Physical AI is now powering the next age of robotics. At this time, it is possible to mechanize tasks that were once too variable or costly, as explored in a paper just published by BCG and the World Economic Forum—in collaboration with dozens of leading private-sector and academic robotics experts. Traditional robots handle repetitive, predictable tasks. With physical AI, robots can see, think, and act—enabling a new level of autonomy and adaptability. These new robots are powered by the convergence of vision systems, AI models, and advanced hardware. They are closing the gap between the digital and physical worlds. Increasingly, robots are trained in virtual environments before reaching the factory floor—making them faster to roll out and more affordable, too. The change is already visible in warehouses and factories. Amazon runs the world’s largest robotics fleet, achieving deliveries 25% faster. Foxconn, the largest global electronics maker, is using AI-powered robots and digital twins to handle precise tasks such as tightening screws—once considered too tricky for machines to handle—cutting costs by 15%.”
