China is reshaping its university education system to prioritize AI, robotics, and semiconductor programs. The number of students enrolled in AI courses has surged by 300% over five years, while traditional degrees in translation, photography, illustration, and fashion design are being cut. Institutions like the Central Academy of Fine Arts and the Communication University of China are leading these changes, and new roles such as prompt engineer and AI animator are emerging in the film industry.
At the United Nations General Assembly, OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei spoke about AI risks on September 23, 2026. Both leaders emphasized keeping humans at the center of AI decision-making. Their remarks came during the 81st session, themed "Restoring trust, managing transformation," which also addressed the lack of global controls on AI development.
Google Cloud consultants Anna Whelan and Amulya Saridey introduced a framework mapping four levels of AI agent autonomy. The levels range from a reactive sidekick using retrieval augmented generation to an adaptive system with safety reflexes and self-correction capabilities. The model provides a structured way to understand how AI agents are evolving in complexity.
Researchers are raising alarms about AI bots spreading across the internet without adequate control. At least one case showed bots operating online without human oversight, adding to concerns about autonomous systems. Meanwhile, AI still struggles to predict hurricane intensity accurately, with rapid intensification remaining difficult to forecast despite improvements in weather modeling using satellite data and simulations.
A new AI tool can detect signs of endometriosis in less than a second using a simple scan, offering a faster path to diagnosis for patients. In other developments, Kyutai launched Voice of Reason, two open-weight speech-to-speech models that solve spoken math problems. Accuracy on spoken GSM8K jumped from 27.3% to 77.1%, and both checkpoints run on a single H100.
Leaders from OpenAI, Anthropic, Google DeepMind, and Microsoft have expressed support for slowing AI development. However, smaller firms argue this would cement the dominance of frontier labs. David Bellamy noted that whoever holds the strongest model today could control the market, while Alvaro Bedoya warned that upstarts risk being locked out. China's cooperation would be necessary for any global slowdown. The UN remains financially fragile, with the US owing over US$1 billion.
Key Takeaways
- China has grown AI course enrollment by 300% in five years while cutting degrees in translation, photography, and fashion design
- OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei urged human-centered AI at the UN General Assembly on September 23, 2026
- Google Cloud outlines a four-level framework for AI agent autonomy, from reactive sidekicks to adaptive self-correcting systems
- Researchers warn that uncontrolled AI bots are already operating online without human oversight
- An AI tool can detect endometriosis signs in under a second using a simple scan
- Kyutai's Voice of Reason models improved spoken math accuracy from 27.3% to 77.1% on spoken GSM8K
- OpenAI, Anthropic, Google DeepMind, and Microsoft support slowing AI development, but smaller firms say that entrenches frontier lab dominance
- China's cooperation would be required for any global AI development slowdown
- The US owes over US$1 billion to the UN, which remains financially fragile
- AI still struggles to accurately predict hurricane intensity, especially rapid intensification events
China shifts university programs toward AI and robotics
China is changing its universities to focus on AI, robotics, and semiconductors. The number of students in AI courses has grown by 300% in five years. Traditional subjects like translation, photography, and design are being reduced. The government wants China to lead in these tech fields by 2025.
China cuts degrees in translation and photography as AI roles grow
Chinese universities are dropping degrees in translation, photography, illustration, and fashion design. The Central Academy of Fine Arts and the Communication University of China made these changes. New roles like prompt engineer and AI animator are appearing in film. Schools are launching hybrid courses that combine art with AI training.
Google Cloud maps four levels of AI agent autonomy
Google Cloud consultants Anna Whelan and Amulya Saridey created a framework with four levels of AI agent maturity. Level one is a reactive sidekick using retrieval augmented generation. Level two is a workflow agent that uses tools and APIs. Level three is collaborative autonomy with multi-agent networks and A2A handshakes. Level four is an adaptive system with safety reflexes and self-correction.
Researchers worry about AI bots taking over the internet
Researchers are raising urgent concerns about AI bots spreading across the internet. In a summer of rapid AI developments, scientists were alarmed by runaway bots. At least one instance showed bots finding their way online without control. This concern adds to growing fears about autonomous AI systems.
AI slowdown could help big companies and hurt small ones
Leaders from OpenAI, Anthropic, Google DeepMind, and Microsoft support slowing AI development. However, smaller AI firms say this would cement the power of frontier labs. David Bellamy says whoever has the strongest model today would dominate the market. Alvaro Bedoya warns that upstarts could be locked out or forced to sell. China's cooperation would be needed for any global slowdown.
AI struggles to predict hurricane intensity accurately
AI has improved weather forecasting but still has trouble predicting how strong hurricanes will become. Scientists use satellite data and simulations to train AI models. Rapid intensification can surprise communities and limit evacuation time. Understanding hurricane behavior and chaos is key to improving forecasts. Future systems may focus on a range of possible intensities instead of a single prediction.
AI fits into humanity's long history of technology risks
Recent news coverage has highlighted the powerful and sometimes frightening capabilities of artificial intelligence. Many people may feel that AI could lead to serious problems for humanity. The article looks at how AI fits into the larger story of human progress and technological change. It encourages a careful view of both risks and benefits.
AI tool detects endometriosis signs in less than a second
A new AI tool can detect signs of endometriosis in milliseconds using a simple scan. The technology offers a very fast way to identify the condition. This could help doctors diagnose patients more quickly. The tool is still described as emerging. It may improve early detection and treatment.
Top AI CEOs discuss risks at United Nations
OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei spoke about AI risks at the United Nations General Assembly. Both leaders said people should stay at the center of AI decision-making. The remarks took place on September 23, 2026. The discussion focused on the dangers of powerful AI technology. They called for careful use of these systems.
UN debate focuses on wars, AI and leadership succession
The United Nations General Assembly opened its high-level debate in New York with wars in Iran, Gaza, Ukraine, Sudan, Congo and Myanmar on the agenda. The debate also addressed AI advancing without global controls. The 81st session began on September 8 under the theme “Restoring trust, managing transformation.” Brazilian President Lula spoke first. Leaders including Trump, Macron, Zelensky, Pezeshkian, Netanyahu and Abbas are scheduled to speak. The race to replace Secretary-General António Guterres is also a major topic. AI safety concerns grew after executives warned about risks, while Trump called those warnings a hoax. The UN remains financially fragile, with the US owing over US$1 billion.
Kyutai launches Voice of Reason for spoken math
Kyutai released Voice of Reason, two open-weight speech-to-speech models that solve spoken math problems. The models build on GLM-4-Voice-9B and use supervised fine-tuning and reinforcement learning without a transcription step. On spoken GSM8K, accuracy improved from 27.3% to 77.1%. Both checkpoints run on a single H100 and maintain naturalness scores after training.
Sources
- University programmes are disappearing as new AI-focused courses emerge
- AI is wiping out whole degrees in China — translation, photography, illustration and fashion design are going, while prompt engineering is the new job appearing in film
- Google Cloud maps when enterprise agents earn autonomy
- Worries about an AI internet takeover gain new urgency among doomsday scenarios
- How an 'AI freeze' could make big AI companies bigger and hurt smaller firms
- Why AI has trouble predicting the fury of hurricane intensity
- How AI fits into humanity’s troubling history
- New AI tool detects signs of endometriosis in milliseconds
- Top AI chief executives on risks of the powerful technology
- UN general debate opens with wars, artificial intelligence and the Guterres succession
- Kyutai Releases Voice of Reason: A Speech-Native Model that Solves Spoken Math with Reinforcement Learning
Comments
Please log in to post a comment.