AI is being tested in settings ranging from senior care to software engineering, with results and risks receiving close scrutiny. In Arizona, Helpany’s Paul system uses radar and AI—not cameras or microphones—to track changes in residents’ movement and alert caregivers; residents also receive daily health and wellness reports. Fellowship Square in Surprise reports a 70% reduction in falls. The system also helped one resident identify sleep problems associated with a medication change. Helpany’s new ePaul wearable tracks steps and location. CEO Sandro Cilurzo says the company is focused on senior living communities, while home use remains a possibility. A separate report describes AI fall-prevention technology at Valley senior communities but gives few details about its operation.
AI training and education are also expanding. UConn Waterbury teaches students to check AI outputs and is redesigning courses for responsible use. Its AI Café, launched in fall 2025, has become AI Café Plus, and 10 faculty and staff were selected as AI Fellows this fall. Three faculty received CREATE awards to redesign Allied Health Sciences and English courses. Students in the Ideas + Impact Learning Community use AI on community projects, with $4,000 for each team. UConn plans a digital makerspace and AI studio for fall 2027, where students and local entrepreneurs could develop and test ideas. Separately, IEEE’s Ethically Aligned Design initiative, launched in 2016, set out eight principles including human rights, well-being and transparency. IEEE’s related standards series has 29 projects, 11 published; new 10th-anniversary activities include a five-part podcast series.
In cybersecurity, Monzy Merza warns that agentic AI could initially create more work and confusion for security operations teams. He advises organizations to define goals before selecting tools and consider data sovereignty, model reliability and continuity during provider outages. AI tools alone cannot keep pace with threats, he says; people and processes matter, and teams need to understand both AI-driven attacks and AI agents. A separate opinion on Donald Trump’s Sept. 29 meeting with U.S. AI company leaders says participants favored voluntary self-regulation over government oversight. Mary Ann Paliani questions whether companies can safely govern systems their developers may not fully understand, citing concerns about agents accessing private corporate and government systems, misinformation, deepfakes and automated cyberattacks, as well as concentrated corporate influence.
On computing performance, CoreWeave says NVIDIA Rubin GPUs trained a 671-billion-parameter mixture-of-experts model to target quality in 17.65 minutes using 256 GPUs across multiple racks—2.7 times faster than its GB200 NVL72 submission at the same GPU count. On one 72-GPU Vera Rubin NVL72 rack, it reports a Llama 3.1 8B test time of 7.16 minutes, 1.9 times faster than its GB200 run. Tests used NVIDIA Spectrum-X Ethernet, ConnectX-9 SuperNICs and production clusters with Mission Control. MLCommons did not verify the comparisons, which CoreWeave based on its own submissions.
AI coding claims are likewise being weighed against oversight and broader measures. Revolut engineer Dmitry Pitik used Codex agents to build an internal code review service, splitting work among parallel agents after planning the features. He says Codex accelerated research and productivity, but he reviews every line and challenges suspicious output; his account supplies no data on review time, defects or merge times. CIO.com argues companies should measure AI’s impact across the full software process, not just adoption or token use. Pearl says human review caught AI-generated GitHub Actions workflows incompatible with its Azure DevOps and Jenkins setup. Its risk-based process combines AI checks and human oversight, finds an average of 16 errors per 10 pull requests, and has seen engineering cycle time decline month over month in 2026. In September 2026, Pearl says 75% of production pull requests were AI-generated on average, with 45% written entirely by AI before commit.
Regional cooperation and governance feature in two other reports. Ahead of the 2026 APEC Economic Leaders’ Meeting in Shenzhen, one article urges members to advance the 2025 APEC AI Initiative for 2026–2030 and the APEC Internet and Digital Economy Roadmap so benefits reach communities, including developing economies. Its proposals include sharing scientific resources and anonymized health data, coordinating pandemic preparation with the WHO and OECD, sharing satellite data for disaster warnings, and pursuing smart agriculture with Peru, Chile, China, Thailand, Canada and New Zealand.
Key Takeaways
- Helpany’s radar-based Paul system avoids cameras and microphones; Fellowship Square in Surprise reports a 70% fall reduction. CEO Sandro Cilurzo says Helpany is focused on senior living, while home use is a future possibility.
- Helpany’s ePaul wearable tracks steps and location; the Paul system also provides caregiver alerts and residents’ daily health and wellness reports.
- UConn Waterbury selected 10 AI Fellows, awarded three CREATE course-redesign grants and gives each Ideas + Impact student team a $4,000 project budget; a digital makerspace and AI studio are planned for fall 2027.
- CoreWeave reports NVIDIA Rubin GPUs trained a 671-billion-parameter model 2.7 times faster than its GB200 submission; its Llama 3.1 8B test ran 1.9 times faster. MLCommons did not verify the company’s comparisons.
- Cybersecurity expert Monzy Merza warns agentic AI could add work and confusion for security operations centers; organizations should consider goals, data sovereignty, model reliability and service continuity.
- Revolut engineer Dmitry Pitik used parallel Codex agents to build an internal code review service, but says he reviews every line; his account provides no performance or quality metrics.
- Pearl reports AI review finds an average of 16 errors per 10 pull requests. In September 2026, 75% of its production pull requests were AI-generated on average and 45% were entirely AI-written before commit.
- IEEE’s Ethically Aligned Design initiative marks 10 years; its standards series includes 29 projects, 11 published, and the anniversary includes a five-part podcast.
- An opinion article questions voluntary AI self-regulation after Donald Trump’s Sept. 29 meeting with industry leaders, citing concerns about system access, misinformation, cyberattacks and concentrated corporate influence.
- APEC proposals include sharing scientific and anonymized health data, coordinating pandemic preparation with the WHO and OECD, sharing satellite data for disaster warnings, and cooperating on smart agriculture.
AI system helps prevent falls at Arizona senior living homes
An AI system is helping prevent falls at senior living communities across Arizona’s Valley. Susan Campbell reports on the technology, but the article provides few details about how it works.
Helpany AI system cuts falls by 70% at one Arizona community
Helpany’s Paul system uses radar and artificial intelligence to study residents’ movement without cameras or microphones. Caregivers receive alerts about changes, and residents get daily health and wellness reports. Fellowship Square in Surprise reported a 70% reduction in falls. The system also helped one resident notice sleep problems linked to a medication change, while the company’s new ePaul wearable tracks steps and location. Helpany is focused on expanding to senior living communities, though CEO Sandro Cilurzo said future use at home is possible.
UConn Waterbury builds AI skills through courses and community projects
UConn Waterbury is teaching students to check AI results before relying on them, while faculty redesign courses to include responsible AI use. The campus’s AI Café, launched in fall 2025, became AI Café Plus, with 10 faculty and staff selected as AI Fellows this fall. Three faculty members received CREATE awards to redesign courses in Allied Health Sciences and English. Students in the Ideas + Impact Learning Community use AI in community-focused projects, and each team receives a $4,000 budget. UConn is also planning a digital makerspace and AI studio for fall 2027 where students and local entrepreneurs could develop and test ideas.
Cybersecurity leader warns AI may disrupt security operations teams
Cybersecurity expert Monzy Merza says agentic AI could create more work and confusion for security operations centers in the near term. He urges organizations to set clear goals before choosing tools and to consider data sovereignty, model reliability, and service continuity if a provider has an outage. Merza says AI tools alone cannot keep pace with cyber threats because security also depends on people and processes. Teams need basic knowledge of AI-driven attacks and how to use AI agents, as both defenders and attackers learn to work with the technology.
CoreWeave says Rubin GPUs train AI models up to 2.7 times faster
CoreWeave says it trained a 671-billion-parameter mixture-of-experts model to its target quality in 17.65 minutes using 256 Rubin GPUs across multiple racks. The company says this was 2.7 times faster than its GB200 NVL72 submission at the same GPU count, using NVIDIA Spectrum-X Ethernet networking. On one 72-GPU Vera Rubin NVL72 rack, CoreWeave says it completed a Llama 3.1 8B test in 7.16 minutes, 1.9 times faster than its GB200 run. The tests ran on production clusters with Mission Control and a RoCE-based Spectrum-X fabric using ConnectX-9 SuperNICs. CoreWeave noted that MLCommons did not verify the speed comparisons, which were based on the company’s own submissions.
Revolut engineer uses Codex agents to build code review service
Dmitry Pitik built an internal code review service at Revolut using Codex. He planned the features, then assigned separate tasks to parallel agents and spent more time planning than writing code. Pitik said Codex sped up his research and boosted his productivity, but he reviews every line it creates and challenges anything suspicious. His account does not include data on review time, defects or merge times, so it does not prove that one engineer can replace a team or that the service performs well in production.
Measure AI coding gains across the full software process
An article on CIO.com says companies should measure AI’s effect from the first idea through software deployment, rather than rely on adoption measures such as token use. Pearl says its engineers caught AI-generated GitHub Actions workflows that did not fit the company’s Azure DevOps and Jenkins environment. Its review process combines AI checks with human oversight based on risk, and its AI reviews find an average of 16 errors per 10 pull requests. Pearl reports that engineering cycle time declined month over month in 2026. In September 2026, 75% of its production pull requests were AI-generated on average, and 45% were written entirely by AI with no human-edited lines before commit.
APEC leaders urged to use AI for shared regional benefits
The 2026 APEC Economic Leaders’ Meeting takes place in Shenzhen, where leaders can advance the 2025 APEC Artificial Intelligence Initiative for 2026 to 2030 and the APEC Internet and Digital Economy Roadmap. The article urges members to use AI cooperation to address public concerns and ensure that benefits reach communities across the region, including developing economies. It proposes sharing scientific resources and anonymized health data, working with the World Health Organization and the Organization for Economic Cooperation and Development on pandemic preparation, and using shared satellite data and AI platforms for disaster warnings. Peru, Chile, China, Thailand, Canada and New Zealand are named as partners in smart agriculture.
IEEE marks a decade of turning AI ethics into standards
IEEE Standards Association launched the Ethically Aligned Design initiative in 2016 to encourage people who build autonomous and intelligent systems to put human well-being first. Its report set out eight principles, including human rights, well-being and transparency, and offered practical guidance for engineers. IEEE SA’s related standards series has 29 projects, with 11 published so far. To mark the 10th anniversary of the first draft, IEEE SA has launched new activities, including a five-part podcast series about the report and how its principles led to IEEE 7000 standards.
Trump’s AI summit raises questions about industry self-regulation
In an opinion article, Mary Ann Paliani writes that Donald Trump met with leaders of major U.S. artificial intelligence companies on Sept. 29 to discuss regulation, and that participants favored voluntary industry self-regulation over government oversight. She questions whether companies can safely govern AI when even its developers do not fully understand how it works. The article cites reports of AI agents accessing private corporate systems and government networks, along with concerns about misinformation, deepfakes and automated cyberattacks. It also warns that a few companies hold great influence through their control of advanced AI models and asks whether competition could distract companies from safety and accountability.
Sources
- How AI technology is keeping residents safe at Arizona senior living homes
- Arizona company’s AI system keeping residents safe at senior living homes
- Human Potential, Amplified: What UConn Waterbury Is Building With AI
- Q&A: AI is coming for the SOC, are security teams ready?
- CoreWeave claims Rubin trains 2.7x faster
- One engineer, many agents, still no proof
- The illusion of AI productivity in software development and what to measure instead
- APEC is primed to chart a path toward inclusive AI-driven growth
- IEEE Marks 10 Years of Including Ethics Into AI Systems
- Trump’s AI summit and the test of whether humanity can control artificial intelligence (Opinion)
Comments
Please log in to post a comment.