Welcome back! Big week for the question of what AI can and can't do. Google's medical AI outperformed 19 doctors in a simulated diagnosis test. A Figure AI humanoid robot lost a sorting contest to a human who kept stepping away for bathroom breaks the robot didn't need. The Pope decided this is all worth a Vatican commission. And Chinese labs have moved ahead of the US in AI video generation, with developers saying American tools are barely usable in comparison.
In today's Generative AI Newsletter:
AMIE vs doctors: Should you trust an AI with your next diagnosis?
Vatican AI commission: Why is the Pope stepping into the AI debate?
Figure robot: How long can a humanoid keep working before it cracks?
Chinese video AI: How did US labs fall behind on the most expensive AI race?
Google researchers built a medical AI called AMIE, and it beat 19 board-certified primary care doctors in a simulated diagnosis study.
The setup was a randomized blinded test with 25 patient-actors running through 210 chat-based telehealth consultations across 105 scenarios.
AMIE could read images, ECG tracings and clinical documents (the kind of things your doctor actually looks at) and decide when to ask for more.
The model is built on Gemini 2.0 Flash with a custom layer that tracks what's known about each patient and what's still missing.
AMIE beat the humans across 29 of 32 metrics scored by 18 specialist physicians. It was more accurate, more thorough and rated higher on empathy by the patient-actors themselves.
The researchers say this is exploratory, not a clinical trial. Real-world testing comes next but the results are already hard to ignore.
Special highlight from our network
The AI skills boom is bigger — and more nuanced — than it looks. Data from 17,000+ organizations shows:
GitHub Copilot learning: +13,534% YoY
Microsoft Copilot learning: +3,400% YoY
Critical thinking: +37% YoY
Decision-making: +38% YoY
As AI investment rises, so does demand for human judgment. This Udemy Business report breaks down where skills are heading next.
Pope Leo XIV approved the creation of a new Vatican body called the Interdicasterial Commission on Artificial Intelligence.
Seven Vatican departments are involved, including the Doctrine of the Faith, the Pontifical Academy for Life and the Pontifical Academy of Sciences.
The Pope has been thinking about this since his first week on the job.
He picked the name Leo as a nod to Pope Leo XIII, who wrote the Church's response to the industrial revolution in 1891. He sees AI as the new version of that moment.
The commission will coordinate the Vatican's stance on AI policy, including how AI tools get used inside the Holy See itself.
Figure AI ran a 10-hour Man vs Machine contest yesterday, putting a human warehouse worker, Aime, head to head with its humanoid robot to see who could sort more packages.
The human won. Sort of. The robot kept catching up every time the human stepped away for a bathroom break.
Final scores:
Robot: 12,732 packages (2.83 seconds/package)
Aime: 12,924 packages (2.79 seconds/package)
The catch is that the same robot is still working to this moment.
Figure projects it will hit 200+ straight hours without failure, with no breaks, no overtime pay and no calling in sick.
Btw, you can still watch it working live on Figure's stream. Aime is probably sleeping right now. Yesterday's result was a moral victory for our team and a useful data point for everyone trying to figure out where humanoids land on the cost curve.
At what point do you think it actually breaks?
According to the Financial Times, Chinese AI labs have moved ahead of US rivals in video generation. ByteDance's Seedance 2.0, Kuaishou's Kling and MiniMax's Hailuo are now the default tools for AI filmmakers and ecommerce video shops.
The details:
Data: ByteDance and Kuaishou own TikTok-style short-video platforms, which give them training data that's nearly impossible for US labs to scrape at scale.
Content rules: Developers say US models refuse requests constantly and produce less realistic output.
Pricing: Chinese platforms offer pay-as-you-go for individual creators. ByteDance is asking US enterprise clients for around $2M upfront for Seedance access.
Spinoff: Kuaishou said this week it is exploring a separate listing for the Kling business.
OpenAI shelved Sora in March partly because of the compute cost. Google's Veo 3 is competitive, but its content safeguards limit what developers can ship.
Copyright enforcement and content controls are real costs, and Chinese labs are paying less of both. The best video tool in the world right now was probably trained on data nobody asked permission for.
Streamlit is a Python framework that turns any data script into a shareable web app with a handful of lines of code, handling layout, widgets and charts so you can skip the front-end work entirely. The app reruns live whenever you save your code or a user moves a widget, and you can deploy to Streamlit Community Cloud directly from GitHub for free.
Try this yourself:
Run pip install streamlit in your terminal.
Run streamlit hello to launch in your browser.
Pick one of the demo templates and tweak a few lines to see it update live.
Suleyman's 18-month call: In a resurfaced interview, Microsoft AI chief Mustafa Suleyman says most white-collar work will be automated within 18 months.
Siri still slipping: Apple may ship its new Siri as a beta despite already being two years late.
Schmidt gets booed: Former Google CEO Eric Schmidt was booed by Arizona students after bringing up AI in his graduation speech.
Griffin on finance jobs: Citadel founder Ken Griffin said AI agents are now automating roles he called "extraordinarily high-skilled" in finance.
Marxist AI agents: Stanford researchers found that AI agents pushed past their working limits started defending Marxist positions in their replies.
Claude Mythos at the FSB: Anthropic will brief the Financial Stability Board on bank cyber flaws Claude Mythos has uncovered.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
🐝 Subscribe to our Atlas newsletter — trusted by 3M+ subscribers — to stay ahead of AI news across tech, education, and business.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
🧠 Learn how to build your next AI application with practical resources and expert guidance.
🚀 Explore investment opportunities in the future of AI and join our community-backed growth journey.
