Researchers say statistical patterns in the starting weights of open AI models may reveal which Nvidia GPU families trained them. Unemployment Labs compared published weights from DeepSeek, Moonshot, MiniMax and Inclusion AI with patterns associated with Nvidia Hopper and Blackwell chip layouts. Their estimates point to training clusters of roughly 3,000 to 4,500 GPUs and 7,000 to 22,000 chips across several models. The researchers stress that the estimates depend on training dates, hardware and software efficiency, and whether models shared clusters. The method cannot establish where chips were used or whether export rules were violated.
AI’s implications also extend beyond computing. Nir Eisikovits, a philosophy professor at UMass Boston, argues that colleges may need to prepare students for meaningful leisure if AI reduces demand for human work. Drawing on Aristotle, he emphasizes learning, relationships, play and attention to art, alongside critical thinking and self-discipline. Stanford plans a winter 2027 course on AI through racial capitalism theory, including discussion of materials used in AI infrastructure such as Zambian copper, Namibian uranium and Congolese cobalt. The course listing provides no syllabus or reading list.
In practical settings, E Tech Group AI solutions director Akshatha Shetty says engineers can use AI to tackle time-consuming tasks, provided they supply clean data and check results carefully. She said proposal work that takes one or two weeks could be cut to two days, and recommends measuring value by time saved. In Mozambique, meanwhile, outdoor classrooms, seasonal rain, weak connectivity and regional and linguistic diversity complicate AI adoption. Offline tools, preloaded content and digital teaching assistants are possible options, but the article stresses that AI should complement—not replace—schools, teachers and basic materials.
Two reports highlight governance and institutional responses. South Korea ranks 10th in a global AI ranking, behind the United States, Singapore and China, which place first, second and seventh; its government says it will continue support for AI and digital transformation. A Denver audit found the city lacks a central inventory of AI systems and has not fully incorporated AI into technology risk policies. Auditors also entered simulated prohibited information into a public AI system without triggering a warning. At Denver International Airport, staff used seven unreviewed or unapproved AI tools more often than the approved Microsoft Copilot platform.
AI is also changing publishing and pharmaceutical outreach. A New York Times report cited in an article on Philip K. Dick’s warnings found 2,168 self-published Amazon books with more than half their content generated by AI, including 970 novels with over 90% AI-written pages. The article also describes disputes involving authors Thélyson Orélien and David Silva Rojas, and argues that Dick’s fiction anticipated concerns about distorted reality and diminished human creativity. In China, Alibaba Health and JD Health have introduced AI tools for doctors. Alibaba Health says 93% of several hundred surveyed doctors use AI, up from about 70% a year earlier. As scrutiny makes sales visits to hospitals harder, drugmakers are exploring promotion through AI apps and physician platforms; Alibaba Health is considering sponsored medical evidence in its Hydrion app but has not begun that work with drugmakers.
Key Takeaways
- Researchers link starting-weight patterns in open models including DeepSeek to Nvidia Hopper and Blackwell GPUs, estimating 7,000–22,000 chips across several models; the method cannot establish chip locations or export violations.
- UMass Boston professor Nir Eisikovits says colleges may need to teach students how to use leisure meaningfully if AI reduces the need for human work.
- Stanford plans a winter 2027 course on AI and racial capitalism, including minerals used in AI infrastructure; no syllabus or reading list is listed.
- E Tech Group’s Akshatha Shetty advises engineers to use AI on time-consuming tasks, provide clean data and review outputs; proposal work could potentially shrink from one or two weeks to two days.
- South Korea ranks 10th globally in AI, while the United States ranks first, Singapore second, China seventh and Japan 19th.
- In Mozambique, limited connectivity and basic school needs constrain AI adoption; offline tools and preloaded content are options, but AI should not replace schools, teachers or materials.
- A cited New York Times report found 2,168 self-published Amazon books with more than half their content AI-generated, including 970 novels with over 90% AI-written pages.
- A Denver audit found gaps in AI oversight, including no central system inventory; airport staff used seven unreviewed or unapproved tools more often than Microsoft Copilot.
- Alibaba Health reports AI use among surveyed Chinese doctors rose to 93% from about 70% a year earlier; drugmakers are exploring promotion through AI and physician platforms.
Researchers trace Chinese AI model weights to Nvidia Blackwell chips
Researchers used statistical patterns in published AI model weights to identify possible Nvidia Blackwell and Hopper chips used during training. They compared weight initialization patterns with results from different GPU layouts and found matches for models including DeepSeek and Kimi. The method can estimate chip families and approximate card counts, but it cannot show where the GPUs were located or prove that export rules were broken. The researchers also said their analysis applies only to models with open weights and depends on the hardware and software used.
Study links Chinese AI models to thousands of Nvidia Blackwell GPUs
Unemployment Labs researchers say the starting weights of models from DeepSeek, Moonshot, MiniMax and Inclusion AI carry patterns linked to Nvidia Blackwell GPUs. They compared published model weights with patterns created by the streaming multiprocessor layouts of Nvidia Hopper and Blackwell chips. Their estimates suggest training clusters of about 3,000 to 4,500 GPUs, with total use across several models estimated at 7,000 to 22,000 chips. The researchers caution that the totals depend on factors such as training dates and GPU efficiency, and the highest estimate assumes the models did not share a cluster.
AI could lead colleges to prepare students for leisure
Nir Eisikovits, a philosophy professor at UMass Boston and director of its Applied Ethics Center, considers how colleges might change if AI reduces the need for human work. He suggests universities could prepare students for meaningful leisure, not only careers, if many jobs disappear. Drawing on Aristotle, he describes leisure as time for learning, relationships, play, and appreciating beauty and great works. He argues that people need exposure to different activities, sustained attention, critical thinking, and self-discipline to use free time well.
Stanford course will study AI through racial capitalism theory
Stanford University plans to offer a winter 2027 course examining artificial intelligence through the idea of racial capitalism. The course will cover scholarly debates and theories linked to Karl Marx and Cedric Robinson, as well as minerals used in AI infrastructure, including copper from Zambia, uranium from Namibia, and cobalt from the Democratic Republic of Congo. Stanford’s 2026 to 2027 listings also include courses on AI, colonialism, decolonization, and African archives. The course page did not provide a syllabus or reading list, and it did not say whether students will study competing economic views.
AI can support engineers, but human review remains essential
Akshatha Shetty, the new director of AI solutions at E Tech Group, told Design News that AI can help engineers work more creatively when companies use it well. She advised engineers to start with tasks that take a long time, provide AI with clean and organized data, and closely check its output. She said people must review AI results rather than trust them blindly, because AI can produce incorrect answers. Shetty also said companies can measure AI’s value by tracking time saved, noting that proposal work that takes one or two weeks could be reduced to two days.
South Korea ranks 10th worldwide in artificial intelligence
South Korea has moved into 10th place in a global ranking of artificial intelligence, according to Yonhap News Agency. The United States ranked first, Singapore second, China seventh and Japan 19th. South Korea’s ministry credited government efforts and the country’s resilience for the progress. It said the government will continue policies supporting artificial intelligence and digital transformation.
AI’s promise meets basic education needs in Mozambique
Thousands of students in Mozambique still study outdoors, and rainy weather from about November to March can disrupt classes and school schedules. Limited connectivity, regional inequality, and the country’s linguistic and cultural diversity create challenges for bringing artificial intelligence into education. Possible options include offline tools, preloaded content, and digital teaching assistants. The article says AI should support, not replace, essential needs such as building schools, training teachers, and providing basic materials. It argues that decisions should respect local conditions and keep students and teachers at the center.
Philip K. Dick’s warnings about AI and the future of literature
The article connects Philip K. Dick’s 1965 novel The Three Stigmata of Palmer Eldritch with current concerns about AI-generated books and writing. It cites a New York Times report that found 2,168 self-published Amazon books had more than half their content generated by AI, including 970 novels with more than 90% AI-written pages. It also describes literary cases involving Thélyson Orélien, removed from the Prix Goncourt over claims of AI use and plagiarism, and David Silva Rojas, accused of using 99% AI in a prize-winning story. The article argues that Dick foresaw risks from technology, including distorted reality and reduced human creativity and feeling.
Denver audit finds gaps in AI safeguards at city and airport
A Denver audit found weaknesses in how the city and Denver International Airport oversee artificial intelligence. Auditors entered simulated prohibited information into a public AI system without triggering a warning, and city staff said a deployment problem kept security updates from reaching some city computers. The city also lacks a central list of AI systems and has not fully added AI to its technology risk policies. At the airport, employees used seven unreviewed or unapproved AI tools more often than the approved Microsoft Copilot platform. Auditor Timothy M. O’Brien said both the city and airport need stronger oversight to make AI use safe, ethical, and secure.
Drugmakers explore AI apps to reach Chinese doctors
Chinese companies including Alibaba Health and JD Health launched AI tools for doctors this year, and drugmakers are exploring ways to use them to reach physicians. Alibaba Health reported that 93% of several hundred surveyed doctors use AI tools, up from about 70% a year earlier. Increased scrutiny of hospital visits by sales staff has made traditional drug promotion harder, leading companies to consider promoted research in AI apps and disease-focused platforms. Alibaba Health is considering placing promoted medical evidence beside answers from its Hydrion app, but has not started working with drugmakers on that plan. Other firms, including PICA Health and Null Hypothesis, have developed services for drugmaker promotion or disease-focused physician platforms.
Sources
- Researchers allege DeepSeek, Kimi, and other Chinese AI models were trained using banned Nvidia Blackwell GPUs — Researchers use new GPU 'fingerprinting' technique to identify the GPUs used to train open models
- DeepSeek And Moonshot AI Models Appear To Carry Hidden NVIDIA Blackwell Fingerprints In Their Starting Weights, Pointing To Training Clusters Of 3,000 To 4,500 GPUs
- Opinion | Because of AI, colleges may train people for leisure
- Stanford to teach AI through 'racial capitalism,' Marxist theory
- AI Won't Replace Engineers, Expert Says
- South Korea advances to 10th place globally in Artificial Intelligence
- Learning under trees in the era of AI: The paradox of education in Mozambique
- How Philip K. Dick Predicted the Future of Artificial Intelligence in literature
- Denver Audit Exposes Major Gaps In AI Security At City Hall And Denver International Airport
- Drugmakers Use China AI Apps to Reach Doctors
Comments
Please log in to post a comment.