Sam Altman: AI Now Smarter Than Top Scientists

Key Takeaways

• Sam Altman says AI is now mostly smarter than top human scientists
• OpenAI claimed its model solved a Navier-Stokes equation, drawing mathematician criticism
• Bret Taylor compares AI job fears to past technological misunderstandings
• Zyphra AI's ZHA method scored 63.4% on long document retrieval vs 58.0% for dense attention
• 67% of security leaders struggle to keep up with AI-related business needs
• 56% report fewer security staff as AI takes over routine cybersecurity tasks
• Amazon Quick unifies structured and unstructured data through a conversational interface
• nCircleTech launched nVibe, an AI app builder for AEC professionals
• Three health systems test AI tools including da Vinci 5 and real-time EEG
• U.S. depends on Taiwan for chip expertise to maintain AI leadership

3 health systems test new AI tools in patient care

Mass General Brigham, Cleveland Clinic, and UH Cleveland Medical Center are exploring new AI uses in healthcare. Mass General Brigham uses algorithms called FAHR-FaceAge and FAHR-FaceSurvival. Cleveland Clinic tests real-time EEG interpretation. UH Cleveland Medical Center surgeons use the da Vinci 5 platform.

Cleveland Clinic and University Hospitals trial clinical AI

Cleveland Clinic and University Hospitals are testing how AI can help in clinical settings. The hospitals use tools like real-time EEG interpretation and the da Vinci 5 surgical platform. Mass General Brigham also uses AI algorithms for patient care. The report was published November 3, 2026 by Joyce Famakinwa.

Security teams struggle as AI changes cybersecurity jobs

67% of security leaders say their teams struggle to keep up with business needs. AI-related risks and identity threats are the top challenges. 56% report fewer security staff as AI takes over routine tasks. Cybersecurity job postings requiring AI skills doubled in G7 countries. Europe is short over 300,000 cybersecurity workers.

Study says AI alignment training may not build true norms

A new paper argues that current AI alignment training cannot create agents that follow norms in all situations. Training that rewards or penalizes behavior only teaches norms as a cost. The paper was posted on arXiv with ID 2611.02162. It says reward-based methods lead to conditional compliance only.

Tech leader says AI job fears are nothing new

Bret Taylor, founder of Sierra and board chair of OpenAI, spoke at Masters of Scale Summit. He said AI-driven job displacement is real but not the end of work. He compared it to Ben Franklin being confused by podcasting. He believes future jobs will be just as surprising.

nCircleTech launches nVibe AI app builder for AEC teams

nCircleTech launched nVibe, an AI-powered platform that lets AEC professionals build applications without coding. Users describe a workflow in plain language, and nVibe creates the app in a sandboxed environment with IT controls. It understands AEC concepts like WBS codes, CSI MasterFormat, RFIs and LoD 100-500. It includes connectors for Autodesk tools, SAP, Procore and other systems.

Amazon Quick unifies structured and unstructured data

Amazon Quick is a tool that combines structured and unstructured data in one place. Users can connect Amazon S3 as a knowledge base, enrich a dataset and build a Space. It supports a custom chat agent that answers business questions across both data types through one conversational interface.

Sam Altman says AI now surpasses top human scientists

OpenAI CEO Sam Altman stated that AI is now mostly smarter than the brightest human scientists. He spoke at a conference about the rapid progress in AI intelligence. OpenAI recently claimed its model solved a Navier-Stokes equation from the Millennium Prize Problems. The claim drew criticism from mathematicians who questioned the rigor of the AI-generated proofs.

Government faces growing AI accountability challenges

A new article warns that government agencies must address accountability as AI agents spread. It highlights risks including over-zealous agents, subverted agents and agent swarms. The article stresses that only real people can be held accountable for actions. It calls for external controls at database and system access points to maintain human oversight.

America's AI future relies on Taiwan's chip expertise

The United States depends on Taiwan for its leadership in artificial intelligence. Taiwan provides critical expertise in chip design and manufacturing for advanced AI systems. The U.S. relies on Taiwan for high-performance computing chips and advanced memory chips. The partnership aims to maintain the global balance of power and prevent China from dominating AI.

Zyphra AI launches new ZHA method for faster AI

Zyphra AI has released a new technology called Zyphra Hierarchical Attention, or ZHA. This method helps large AI models process very long texts more efficiently. The system adds only 81,920 trainable parameters to the Qwen3.6-35B-A3B model while keeping the full memory cache. Tests show ZHA performs within a few points of standard dense attention on coding and reasoning tasks. On long document retrieval benchmarks, the deep tree version of ZHA scored 63.4 percent compared to 58.0 percent for dense attention. The technology works by using a tree structure to find relevant information quickly without needing new optimized software kernels.

Sources

• How 3 health systems are exploring new AI use cases
• How Cleveland Clinic, University Hospitals are testing clinical AI
• 67% of security leaders say their teams struggle to meet business needs
• Rewarded for Not Getting Caught: Why Alignment Training Selects for Conditional Compliance and Cannot Instill Norms
• Will AI kill jobs? One tech leader says Ben Franklin would be just as confused by podcasting
• nCircleTech launches nVibe AI app builder
• Bring structured and unstructured data together with Amazon Quick
• Sam Altman Crows That AI Is Now Smarter Than the Brightest Human Scientists
• Government’s next AI challenge: Accountability
• America’s AI Ambitions Depend on Taiwan
• Zyphra AI Introduces Zyphra Hierarchical Attention (ZHA): A Path to True Subquadratic Attention