A doctor in Oxfordshire used AI to search medical journals and discover that his mother's Parkinson's seizures were likely caused by a vitamin deficiency linked to her medication. After doctors administered the vitamin dose, the seizures appeared to stop. Clinicians caution that clinical trials are still needed and patients should not alter medication without medical guidance.
Meanwhile, AI is creating new challenges in everyday life. A TIAA survey finds that 53% of Americans worry about running out of money in retirement. New longevity research shows many people live well past age 78, sometimes into their 90s. The rise of AI and GLP-1 drugs that may extend lifespans adds further uncertainty to retirement planning for workers who still use outdated life expectancy figures.
On the military front, a CNN report revealed that US officials nearly acted on bad intelligence generated by a chatbot. The AI report contained incorrect information about a vessel's contents before a potential military action. Experts warn that the real danger of AI is not superintelligence but chatbot errors in high-stakes decisions like warfare.
In the corporate world, a student at HubSpot used large language models as judges to evaluate AI agent performance during a six-month co-op. The LLM judge scored outputs on accuracy and safety, helping the company identify which AI tools worked and which needed improvement. Separately, NVIDIA released DIN Deploy samples enabling developers to build local AI apps using C++ and TensorRT RTX, supporting models like OpenAI Whisper and Meta SAM 2.1 with major speed gains on GPU.
In Brazil, researchers created SUNTInsight, an agentic AI system that helps transit managers in Salvador analyze bus data in natural language. The system processed data on about 700,000 passengers and 2,000 vehicles, revealing that demand on Line 1637 was concentrated in specific neighborhoods, leading to more targeted solutions like short turns instead of adding buses across the route.
Politically, Sen. Mark Warner vows to press the Artificial Intelligence Risk Management and Security Act of 2026 after Sen. Ted Cruz blocked it. Warner expects Congress to be out of session until after the Nov. 3 midterm elections. Harvard also launched at least five new faculty searches focused on AI across its Division of Science, spanning eight departments. MIT developed InstructMesh, a tool that lets users repair and edit AI-generated 3D models for fabrication without extensive CAD knowledge. South Park's 29th season episode mocks billionaires and AI datacentres, with Butters complaining about a datacentre in his backyard that uses too much electricity.
Key Takeaways
- A doctor used AI to identify a vitamin deficiency causing his mother's Parkinson's seizures, stopping them with a vitamin dose, though clinical trials are still needed.
- 53% of Americans worry about running out of money in retirement as AI and longevity drugs extend lifespans beyond traditional planning assumptions.
- US military officials nearly acted on bad intelligence generated by a chatbot, highlighting the danger of AI errors in high-stakes decisions.
- HubSpot used LLMs as judges to evaluate AI agent performance, scoring outputs on accuracy and safety during a student co-op program.
- NVIDIA released DIN Deploy samples enabling local C++ AI apps using OpenAI Whisper and Meta SAM 2.1 with GPU speed gains up to 206.41x.
- Researchers in Salvador, Brazil created SUNTInsight, an agentic AI system that analyzed data on 700,000 passengers to improve transit decisions.
- Sen. Mark Warner will push the AI Risk Management and Security Act of 2026 after Sen. Ted Cruz blocked it, with Congress out until after Nov. 3 midterms.
- Harvard launched at least five faculty searches focused on AI across eight departments in its Division of Science.
- MIT developed InstructMesh, a tool allowing users to repair and edit AI-generated 3D models for fabrication without extensive CAD knowledge.
- South Park's 29th season episode mocks billionaires and AI datacentres, featuring Butters complaining about a datacentre in his backyard.
Doctor uses AI to help save mother from Parkinson's seizures
Ed Patrick, an anaesthetist from Oxfordshire, used AI to search medical journal articles while his mother was in hospital. He found that a vitamin deficiency linked to her Parkinson's medication may have caused her seizures. Staff gave her a vitamin dose that appeared to stop the seizures. Doctors say clinical trials are still needed and patients should not change medication without medical advice.
Doctor uses AI to help save mother from Parkinson's seizures
Ed Patrick, an anaesthetist from Oxfordshire, used AI to search medical journal articles while his mother was in hospital. He found that a vitamin deficiency linked to her Parkinson's medication may have caused her seizures. Staff gave her a vitamin dose that appeared to stop the seizures. Doctors say clinical trials are still needed and patients should not change medication without medical advice.
AI and longevity changes make retirement planning harder
A TIAA survey finds that 53 percent of Americans worry about running out of money in retirement. New research says people who reach age 65 often live much longer than they expect, sometimes into their 90s. AI and GLP-1 drugs that may extend lifespans add more uncertainty to retirement planning. Many workers still plan based on an average life expectancy of about 78 years.
Student uses LLMs to judge AI model quality at HubSpot
Ryan Amiri, a computer science student, completed a six-month co-op at HubSpot. He used large language models as a judge to test how well AI agents performed tasks like fetching videos or reading emails. The LLM judge scored outputs using clear criteria such as accuracy and safety. His work helped HubSpot understand which AI tools worked best and which needed improvement.
Error-prone AI nearly sparked war between US and China
A CNN report revealed that US military officials nearly acted on bad intelligence generated by a chatbot. The AI report contained wrong information about a vessel's contents before a potential military action. Experts say this shows the real danger of AI is not superintelligence but errors from chatbots. They warn against relying on these tools for high-stakes decisions like warfare.
Agentic AI System SUNTInsight Improves Transit Decisions in Brazil
Researchers in Salvador, Brazil created SUNTInsight, an agentic AI system that helps transit managers analyze bus data using natural language. The system turned data on about 700,000 passengers, 2,000 vehicles, 400 lines and 3,000 stops into targeted operational insights. A test on Line 1637 showed demand was concentrated in specific neighborhoods rather than spread evenly. This finding led to more focused solutions like short turns instead of adding buses across the whole route.
Warner Vows to Press AI Safety Bill After Cruz Block
Sen. Mark Warner, a Democrat from Virginia, said he will keep pushing the Artificial Intelligence Risk Management and Security Act of 2026. The bill was blocked this week when Sen. Ted Cruz, a Republican from Texas, objected to it. Warner expects Congress to be out of session until after the Nov. 3 midterm elections.
NVIDIA DIN Deploy Samples Enable Local C++ AI Apps
NVIDIA released DIN Deploy samples that help developers build local AI applications using C++ and TensorRT RTX. The samples use ONNX Runtime to run models like OpenAI Whisper for speech recognition and Meta SAM 2.1 for image segmentation. Performance tests on DGX Spark showed major speed gains on GPU versus CPU, such as 206.41x faster for Parakeet TDT. The repository supports Windows, Linux and Arm64 platforms.
MIT InstructMesh Tool Repairs AI-Generated 3D Models
MIT developed InstructMesh, a generative AI tool that lets users repair and edit AI-generated 3D models for fabrication. The tool allows users to prompt the system to create a 3D model and then edit specific parts without needing extensive CAD knowledge. InstructMesh makes it easier to 3D print working real-world items from AI designs.
Harvard Launches Five AI-Related Faculty Searches
Harvard's Faculty of Arts and Sciences launched at least five new faculty searches focused on artificial intelligence across its Division of Science. The searches span at least eight departments and seek scientists pushing the limits of AI applications in research. Dean of Science David T. Johnston wrote that the hiring process would look different from past years. One search already posted seeks a professor at the interface of AI and life sciences.
South Park mocks billionaires and AI datacentres in new episode
South Park's 29th season episode looks at the gap between billionaires and regular people. Butters complains about a datacentre in his backyard that uses too much electricity. The billionaires are shown as tiny talking characters living in a place called Billionaire Weenietown. The episode shows how Parker and Stone view the wealthy elite. The creators themselves became billionaires through a $1.5bn deal with Paramount. Butters ultimately sends the billionaires away to Nashville.
Sources
- Son who used AI to help save mum's life hopes case can help other Parkinson's patients
- Son who used AI to help save mum's life hopes case offers Parkinson's clues
- Clients fear outliving savings as AI and longevity upend retirement math
- Finding the right AI models for specific tasks using LLMs as a ‘judge’
- Forget ‘superintelligence’: error-prone AI nearly sparked world war three this month
- Agentic AI Turns Transit Data Into More Targeted Decisions
- Warner vows to keep pushing AI safety bill after Republican objection blocks it
- Build Local AI Apps with C++ and NVIDIA TensorRT RTX Samples
- New tool lets users repair AI-generated 3D models, then fabricate them just the way they want
- Harvard Science Division Launches Five AI-Related Faculty Searches
- South Park has AI datacentres, sentient penises and a billionaire problem – in more ways than one
Comments
Please log in to post a comment.