A former ByteDance engineer turned Peking University professor, Zhang Chi, argues that China's AI industry is falling further behind the US, citing slower development cycles, weaker infrastructure, and lower quality data. He claims Chinese models perform well on tests but fail in real-world use. This contrasts with views from tech leaders like Elon Musk and Jensen Huang, who believe China is catching up or could even pull ahead.
Meanwhile, a reporter discovered that ChatGPT falsely identified comedian Don Rickles as the male celebrity who texted Lena Dunham, as described in her memoir. The AI confidently stated this despite Rickles being 85 at the time, raising questions about trusting AI for accurate information. This incident highlights the ongoing challenges with AI reliability.
Oxford professor Carl-Benedikt Frey suggests the era of scaling AI with more data and computing power is ending, and new ideas are needed for AI to reach its potential. He notes that institutions and incentives must also change to realize AI's productivity benefits, and the future may involve small language models or a mix of different approaches.
Security researchers at Mozilla tested Anthropic's new Mythos AI system and were impressed by its abilities, noting it elevated AI from a competent software engineer to a world-class security engineer. This has drawn attention in Washington due to potential cybersecurity risks and hacking concerns.
Baltimore Inspector General Isabel Cumming faced backlash after sharing a YouTube video with an AI-generated image of Mayor Brandon Scott surrounded by money and smoking a cigar. The mayor's chief of staff called the image deeply inappropriate and racist. Cumming apologized and removed the post, though some officials say this incident should not distract from her work uncovering fraud.
AI has the potential to fundamentally change corporate structures by replacing middle management roles, which were first created over a century ago by the railroad industry. The top 10 public companies now have market values larger than the combined GDP of G7 nations excluding the US, mirroring historical changes but on a much larger scale.
D-ID is launching training videos with built-in AI agents that users can talk to, aiming to make corporate training more engaging. Companies spend about $400 billion annually on employee training, but 74 percent of senior leaders say their organizations are not keeping up with skill demands. Traditional training videos have not changed much in 30 years.
A study by researchers David T. Marshall and Tim Pressley found that AI does not directly reduce teacher workload or improve well-being. Instead, teachers who felt more confident using AI also felt more capable of engaging students, leading to lower perceived workload and less anxiety. School leaders should focus on AI as a tool for engagement, not just productivity.
Google has launched three AI-powered smart glass models on the Android XR platform to compete with Meta. The Gemini Audio Frames are entry-level glasses with voice assistant, the Gemini Display Edition has a microLED heads-up display, and Project Aura is a developer kit with binocular displays. The glasses use the Gemini 2.5 Pro AI system and Project Astra vision system for real-time object recognition.
Visually impaired runners are using AI-powered smart glasses to train for the London Marathon. Tilly Dowler, who has Stargardt disease and about 10 percent useful vision, uses Oakley smart glasses with Meta AI to get live cues about landmarks and distance. Sha Khan, who lost 90 percent of his vision, says the glasses are a part of him now. However, large crowds at marathons can weaken mobile signals and affect reliability.
A new backgrounder from SIPRI maps the military AI industry to help policymakers understand the different companies and products involved. The AI industry includes hardware, software, and application layers. The bottom layer includes chips and cloud infrastructure, the middle layer includes foundation models like OpenAI's GPT and Anthropic's Claude, and the top layer integrates AI into military systems. Understanding this stack is important for arms control discussions.
Key Takeaways
- Former ByteDance engineer Zhang Chi says China's AI industry is falling further behind the US due to slower development, weaker infrastructure, and lower quality data.
- ChatGPT falsely claimed comedian Don Rickles texted Lena Dunham, highlighting AI reliability issues.
- Oxford professor Carl-Benedikt Frey says the era of scaling AI with more data and computing power is ending, and new ideas are needed.
- Anthropic's Mythos AI tool impressed security researchers at Mozilla, elevating AI to a world-class security engineer, raising cybersecurity concerns in Washington.
- Baltimore Inspector General Isabel Cumming faced backlash for sharing an AI-generated image of the mayor that was called deeply inappropriate and racist.
- AI could replace middle management roles, which were first created over a century ago by the railroad industry.
- D-ID is launching interactive AI agents for corporate training videos, aiming to make training more engaging.
- A study found AI helps teachers when it boosts their confidence, not just saves time, leading to lower perceived workload and less anxiety.
- Google launched three AI smart glass models on Android XR to compete with Meta, using Gemini 2.5 Pro and Project Astra.
- Visually impaired runners are using AI-powered smart glasses with Meta AI to train for the London Marathon, though large crowds can affect reliability.
Ex-ByteDance engineer says China AI gap with US is growing
Zhang Chi, a former ByteDance engineer and now a professor at Peking University, says China's AI industry is falling further behind the US. He argues that Chinese models perform well on tests but fail in real-world use. Zhang points to slower development cycles, weaker infrastructure, and lower quality data as key problems. However, tech leaders like Elon Musk and Jensen Huang believe China is catching up or could even pull ahead.
ChatGPT falsely said Don Rickles texted Lena Dunham
A reporter used ChatGPT to solve a blind item from Lena Dunham's memoir about a male celebrity who texted her. ChatGPT confidently said it was comedian Don Rickles, who was 85 at the time. The reporter found this unlikely based on Dunham's description. The incident raises questions about trusting AI for accurate information.
Oxford professor says AI needs new ideas beyond scaling
Oxford professor Carl-Benedikt Frey says the era of scaling AI with more data and computing power is ending. He believes new ideas are needed for AI to reach its potential. Frey notes that institutions and incentives must also change to realize AI's productivity benefits. He suggests the future may involve small language models or a mix of different AI approaches.
Anthropic's Mythos AI tool raises cybersecurity concerns in Washington
Security researchers at Mozilla tested Anthropic's new Mythos AI system on their code and were impressed by its abilities. The tool elevated AI from a competent software engineer to a world-class security engineer. This has drawn attention in Washington due to potential cybersecurity risks and hacking concerns.
Baltimore inspector general faces backlash over AI photo of mayor
Baltimore Inspector General Isabel Cumming shared a YouTube video with an AI-generated image of Mayor Brandon Scott surrounded by money and smoking a cigar. The mayor's chief of staff called the image deeply inappropriate and racist. Cumming apologized and removed the post. Some officials say this incident should not distract from the inspector general's work uncovering fraud in the city.
AI could replace middle management roles in companies
AI has the potential to fundamentally change corporate structures by replacing middle management roles. These roles were first created over a century ago by the railroad industry. The top 10 public companies now have market values larger than the combined GDP of G7 nations excluding the US. This shift mirrors historical changes but on a much larger scale.
D-ID launches interactive AI agents for corporate training videos
A company called D-ID is launching training videos with built-in AI agents that users can talk to. Companies spend about $400 billion annually on employee training, but 74 percent of senior leaders say their organizations are not keeping up with skill demands. Traditional training videos have not changed much in 30 years. D-ID's new approach aims to make training more engaging and effective.
Study shows AI helps teachers when it boosts confidence not just saves time
Researchers David T. Marshall and Tim Pressley studied how AI affects teacher well-being. They found that AI does not directly reduce workload or improve well-being. Instead, teachers who felt more confident using AI also felt more capable of engaging students. This confidence led to lower perceived workload and less anxiety. School leaders should focus on AI as a tool for engagement, not just productivity.
Google launches three AI smart glass models to compete with Meta
Google has launched AI-powered smart glasses on the Android XR platform in three models. The Gemini Audio Frames are entry-level glasses with voice assistant. The Gemini Display Edition has a microLED heads-up display for navigation and notifications. Project Aura is a developer kit with binocular displays for immersive apps. The glasses use the Gemini 2.5 Pro AI system and Project Astra vision system for real-time object recognition.
AI smart glasses help visually impaired runners in London Marathon
Visually impaired runners are using AI-powered smart glasses to train for the London Marathon. Tilly Dowler, who has Stargardt disease and about 10 percent useful vision, uses Oakley smart glasses with Meta AI to get live cues about landmarks and distance. Sha Khan, who lost 90 percent of his vision, says the glasses are a part of him now. The glasses use a front camera and AI to describe surroundings through speakers. However, large crowds at marathons can weaken mobile signals and affect reliability.
Mapping the military AI industry reveals complex supply chain
A new backgrounder from SIPRI maps the military AI industry to help policymakers understand the different companies and products involved. The AI industry is not a single sector but includes hardware, software, and application layers. The bottom layer includes chips and cloud infrastructure. The middle layer includes foundation models like OpenAI's GPT and Anthropic's Claude. The top layer integrates AI into military systems. Understanding this stack is important for arms control discussions.
Sources
- Ex-ByteDance engineer says China's AI is 'far behind' the US
- Lena Dunham left a clue, so AI told me: It was Don Rickles who texted.
- An Oxford professor on AI’s potential for humanity
- How a shadow AI tool transfixed Washington
- Baltimore leaders react as inspector general faces fallout after sharing AI-generated photo of mayor
- AI Could Reshape Corporate Structure
- Corporate Training Videos Are the Worst. Can Interactive AI Agents Make Them Fun?
- We Studied How AI Shapes Teachers’ Well-Being. Here’s What We Found (Opinion)
- Google’s New AI Smart Glasses Are Leaving Meta Behind
- AI smart glasses are helping visually impaired runners take on the London Marathon
- Mapping the military AI industry
Comments
Please log in to post a comment.