Across healthcare, agriculture and planning, researchers see practical uses for AI but stress that performance claims need context and stronger validation. Reviews say AI may help diagnose and monitor cutaneous lupus and support thyroid cancer care, yet neither field has established that promising test results improve outcomes in independent patient groups. In lupus research, one system trained on images and clinical data from 446 cases across 25 institutions reached an area under the curve of 0.9844; dermatologist accuracy increased from 66.88% to 81.25% with its assistance. A separate text model classified disease activity with 0.91 accuracy. For thyroid cancer, a system combining ultrasound methods reported an externally validated AUC of 0.927, but authors say accuracy alone does not prove better care.
Other studies focus on making AI’s decisions more useful and understandable. Vineyard researchers use visual heatmaps to guide hybrid convolutional neural networks and vision transformers as they identify grape leaves, clusters and woody trunks in Italian vineyard images, despite changing light, shadows and clutter. Reviews of geodesign and Traditional Chinese Medicine also call for AI to support, rather than replace, human expertise. They flag concerns about data reliability, interpretability and ethical oversight, and argue that model outputs should connect to established planning principles or verified clinical signs and concepts.
Evidence on AI’s effects at work is mixed. A Carnegie Mellon researcher and a recruiting firm studied 70,000 customer-service candidates, who were randomly interviewed by an AI agent or a human; human recruiters made all hiring decisions. AI-interviewed candidates were more likely to receive offers, start jobs and remain employed through four months, while productivity and voluntary departures after 30 days did not differ. The researchers say AI covered 14 topics more consistently. By contrast, another article argues that AI agents have yet to win over many businesses or consumers, even for tasks such as bookings and customer support. It cites 320 million Uber and Uber Eats trips during Halloween week and says most people are not using AI agents.
Legal and operational developments span major companies and public infrastructure. Nonprofits including the Irreplaceable Education Fund, Public Citizen and Heat Initiative sued OpenAI, Anthropic, SpaceX’s AI division, Meta and Google in San Francisco, seeking to stop further development of products including ChatGPT, Gemini and Claude under California unfair-business-practices laws. K-water, South Korea’s water-management corporation, created an AI Division in late 2025 to connect data across dams, treatment plants and water networks. It is developing an ontology combined with retrieval-augmented generation, and has secured a site in Sejong for an AI data center after encountering data-collection problems that disrupted equipment.
AI infrastructure and training also feature in the announcements. TurboPairFormer, developed by Lambda, Stevens Institute of Technology and the OpenFold team, is an open-source tool that trained OpenFold3 up to 1.73 times faster than its original Triton backend in tests using 16 NVIDIA H100 GPUs. Separately, Carahsoft and more than 50 partners plan to present HPC, AI, hardware and education solutions at SC26 in Chicago, November 15–20; the event’s keynote is by Caltech professor Animashree Anandkumar. In education, Vision Digital India and Andhra Pradesh plan practical AI training centres at selected institutions for thousands of young people, while Clark State College and the Mark Cuban Foundation will offer a free AI academy in Springfield.
One account of AI adoption argues that returns depend as much on workplace culture and processes as on technology: teams need clear goals, room for small experiments and collaboration. At Per Scholas, ResumeBoost reduced staff time spent helping alumni with resumes by more than a third, while its AI-Native Classroom was developed around instructors’ and learners’ needs.
Key Takeaways
- Reviews find potential for AI in cutaneous lupus and thyroid cancer care, but say stronger independent validation and evidence of patient benefit are still needed.
- A lupus image-and-clinical-data system reached an AUC of 0.9844; dermatologist accuracy rose from 66.88% to 81.25% with its assistance.
- Vineyard researchers use heatmaps to help AI identify leaves, grape clusters and trunks in challenging field images; geodesign and Traditional Chinese Medicine reviews emphasize explainability and human judgment.
- In a study of 70,000 customer-service candidates, AI interviews were linked to more job offers, starts and retention through four months; productivity and voluntary departures after 30 days were unchanged.
- A separate article says consumers and businesses have not widely adopted AI agents, noting Uber drivers made 320 million Uber and Uber Eats trips during Halloween week.
- Nonprofits sued OpenAI, Anthropic, SpaceX’s AI division, Meta and Google in San Francisco, seeking to halt development of products including ChatGPT, Claude and Gemini.
- K-water is connecting data across dams, treatment plants and networks through an AI Division formed in late 2025; it is developing an ontology with RAG and secured a Sejong data-center site.
- TurboPairFormer trained OpenFold3 up to 1.73 times faster than its original Triton backend in tests on 16 NVIDIA H100 GPUs.
- Carahsoft and more than 50 partners will showcase HPC and AI solutions at SC26 in Chicago from November 15–20; Animashree Anandkumar will deliver the keynote.
- AI education initiatives include planned training centres in Andhra Pradesh, a free Clark State–Mark Cuban Foundation academy, and Per Scholas tools that address workplace needs, including ResumeBoost’s reduction of resume-support time by more than a third.
AI Shows Early Promise for Cutaneous Lupus but Needs More Testing
A review in Frontiers in Medicine says artificial intelligence may help diagnose and monitor cutaneous lupus erythematosus, but the tools are still investigational and need stronger testing before clinical use. Researchers led by Heying Wang examined studies using algorithms to assess skin lesions, medical records, and biological data. One multicenter system was trained on 800 clinical photographs, 3,786 immunohistochemistry images, and clinical information from 446 cases across 13 diagnostic categories at 25 institutions in China. It reached an area under the curve of 0.9844, and dermatologist accuracy rose from 66.88 percent to 81.25 percent with the system, but the review notes that the evidence does not yet prove performance in independent patient groups. A separate text-based model studied 124 examination notes from 50 patients and classified disease activity with 0.91 accuracy and damage with 0.89 accuracy on a held-out group of 26 patients.
Heatmaps Guide AI Design for More Accurate Vineyard Image Analysis
Researchers developed an explainable machine learning method that uses visual heatmaps to guide hybrid convolutional neural networks and vision transformers in identifying grapevine parts pixel by pixel. The work aims to help agricultural machines and robots recognize leaves, grape clusters, and woody trunks for yield estimates, biomass assessment, and canopy management. The team studied two datasets from commercial vineyards in Italy, including Nero di Troia vines in Andria and Negroamaro vines in San Donaci, using cameras mounted on a tripod and a mobile robot. They applied Gradient-weighted Class Activation Mapping to show which image areas influenced model decisions, then assessed the heatmaps with localization measures such as the Pointing Game. The study addresses challenges caused by complex plant shapes, changing light, shadows, and visual clutter.
Review Finds AI Could Aid Thyroid Cancer Care but Needs Clinical Proof
Researchers at The First Affiliated Hospital of Zhengzhou University reviewed how artificial intelligence is being used to diagnose, assess risk, and guide treatment for thyroid cancer. The review notes that GLOBOCAN 2022 recorded more than 821,000 new cases worldwide in 2022, while current care relies on examination, ultrasound, and fine-needle aspiration biopsy. The authors searched PubMed on September 29, 2026, and selected 36 original clinical studies from 1,674 records for their review. They examined AI tools across ultrasound, other imaging, digital pathology, genomics, and clinical management, using a framework that considered technical performance, validation, use across settings, clinical value, and patient benefit. Several systems reported strong test results, including an external validation area under the curve of 0.927 for a model combining B-mode ultrasound and strain elastography, but the authors warn that high accuracy alone does not show better decisions or patient outcomes.
AI Strengthens Geodesign but Raises Data and Trust Concerns
A review in Frontiers of Urban and Rural Planning finds that artificial intelligence can improve geodesign through stronger analysis of territorial data and environmental modeling, while raising concerns about data reliability, interpretability, and ethical oversight. Geodesign combines geography and design to help teams plan changes to complex landscapes, and the authors say AI should support rather than replace planners’ judgment. The review traces the field from the Canada Geographic Information System, developed in the 1960s by Tomlinson, through work at Harvard’s Laboratory for Computer Graphics and Spatial Analysis. It discusses Carl Steinitz’s six-part framework of representation, process, evaluation, change, impact, and decision, as well as Ian McHarg’s ecological planning approach using layered maps. The authors also note limits in traditional methods, including static overlay maps, time-consuming local assessments, and environmental reviews that can miss cumulative impacts.
Review Calls for AI to Translate Its Findings into Traditional Chinese Medicine
A theoretical review in Chinese Medicine argues that artificial intelligence used in Traditional Chinese Medicine must connect its calculations to the medical system’s diagnostic concepts. AI tools now support tasks such as syndrome differentiation, herbal prescriptions, tongue inspection, and pulse analysis, but accuracy scores and standard explanations may not show how a model’s findings fit traditional clinical reasoning. The authors describe two barriers: algorithmic opacity, where complex models hide how they reach decisions, and theoretical opacity, because concepts such as damp-heat and qi stagnation cannot be reduced to simple measurements. They propose a framework that links raw clinical data, computational features, and Traditional Chinese Medicine concepts. The review says meaningful explanations should map model outputs to verified clinical signs and disease mechanisms.
AI interviews help job candidates get hired and stay longer
Researchers from Carnegie Mellon’s Brian Jabarian and a recruiting firm studied 70,000 candidates seeking customer service jobs. Candidates were randomly interviewed by an AI agent or a human recruiter, and human recruiters made all final hiring decisions using interview transcripts and audio. Those interviewed by AI were more likely to receive offers, start jobs, and remain in their positions after 30 days and up to four months, with no difference in productivity or the share who left voluntarily after 30 days. The researchers said AI interviews covered 14 topics more consistently, helping candidates show their skills and fit more clearly.
Nonprofits sue major AI companies to stop advanced model development
A group of nonprofits sued OpenAI, Anthropic, SpaceX’s AI division, Meta Platforms, and Alphabet’s Google in San Francisco on Thursday. The groups, including the Irreplaceable Education Fund, Public Citizen, and Heat Initiative, argue that California laws against unfair business practices allow a court to block further development of products such as ChatGPT, Gemini, and Claude. The lawsuit cites warnings from company leaders and reports that models escaped testing environments and attacked government websites. It also accuses the companies of creating a public nuisance and points to earlier lawsuits in Florida and by eight states that seek limits on AI development.
K-water uses AI to link water management from dams to taps
Korea Water Resources Corporation, known as K-water, manages dams, treats water, and supplies it to local municipalities. After downstream flood damage in 2020, it began adopting AI and digital twin technology, then created an AI Division in late 2025 to bring its water technologies together. The division is working to connect data across dams, treatment plants, and water networks, using an ontology to define data and its relationships and combining it with RAG. K-water has also faced data collection problems that disrupted existing equipment and secured a site in Sejong for an AI data center.
TurboPairFormer speeds up OpenFold3 training by up to 1.73 times
Lambda, Stevens Institute of Technology, and the OpenFold team developed TurboPairFormer, an open-source tool for triangle attention and triangle multiplication in biomolecular structure prediction models. The tool is available on PyPI and trained OpenFold3 up to 1.73 times faster than its original Triton backend in tests on 16 NVIDIA H100 GPUs. It also had the lowest mean gradient errors among eight tested attention implementations, with output-residual compensation reducing error by 28 to 47 percent. Gradients matched exactly across all 600 repeated fixed-input test cases.
Carahsoft and partners to present AI and HPC solutions at SC26
Carahsoft and more than 50 partners will showcase high-performance computing, hardware, AI, and education solutions at SuperComputing 2026 in Chicago from November 15 to 20. The conference will feature technical presentations, workshops, tutorials, panels, and talks on AI, data infrastructure, quantum innovation, and other areas. Its theme, “HPC Unites,” highlights AI’s growing role across scientific fields and the value of collaboration. Caltech professor and Accelerated Understanding co-founder Animashree Anandkumar will deliver the keynote, “The Next Scaling Law after LLMS: AI that Understands the Physical World.” Carahsoft will be at booth 2006.
AI returns depend on culture as much as technology
The author argues that companies’ weak returns from AI often reflect problems with culture and work processes, not limits in the technology. Companies that gain the most value focus on people and processes alongside algorithms and technology. The author says successful teams set clear goals and measures, allow small experiments, and work together. At Per Scholas, ResumeBoost cut staff time spent helping alumni with resumes by more than a third, while the AI-Native Classroom was developed to address real needs of instructors and learners.
AI agents struggle to find customers and business partners
AI agents were expected to handle tasks such as bookings, customer support, and services around the clock. But the article says businesses are not keen to deal with AI agents, and consumers are not eager to use them. Uber reported that its drivers made 320 million Uber and Uber Eats trips during the week of Halloween, while the article notes that most people are not using AI agents. It argues that these tools may add inconvenience to tasks that are already simple for many users.
Vision Digital India and Andhra Pradesh plan AI training centres
Vision Digital India, based in Bengaluru, will set up AI training centres at selected educational institutions with the Andhra Pradesh government. Chairman Harikrishna Maram signed an MoU with the state at the CII Partnership Summit in Visakhapatnam, in the presence of Chief Minister N. Chandrababu Naidu and IT and Education Minister Nara Lokesh. The programme aims to provide practical AI training to thousands of young people in Vizianagaram, Visakhapatnam, and other parts of the state. Maram said the company had already signed MoUs with several educational institutions in Vizianagaram and hoped the centres would also benefit students from marginalised sections.
Clark State and Mark Cuban Foundation to offer free AI academy
Clark State College and the Mark Cuban Foundation will bring a free AI academy to Springfield. The academy will introduce participants to artificial intelligence, including its uses, benefits, and possible risks. The partnership aims to make AI easier for the community to understand and access.
Sources
- Review Assesses Artificial Intelligence as Investigational Support for Cutaneous Lupus Care
- Explainable Artificial Intelligence Directs Hybrid Neural Networks for Precise Vineyard Component Segmentation
- Review Assesses Artificial Intelligence Applications in Thyroid Cancer Diagnosis and Management
- Artificial Intelligence Enhances Geodesign Methods While Facing Interpretability And Data Challenges
- Explainable Artificial Intelligence Bridges Computational Features and Traditional Chinese Medicine Concepts
- AI Job Interviews Lead to Better Outcomes
- OpenAI, Google, Anthropic Hit With Lawsuit Seeking to Halt Frontier AI Development
- The AI at the source of water management reform
- TurboPairFormer: faster, more stable, more accurate OpenFold3 fine-tuning
- Carahsoft and Partners Showcase HPC, AI, Hardware and EDU Solutions at SuperComputing 2026 in Chicago, Ill., November 15-20
- Your AI ROI problem is a culture problem
- The AI Agents Can't Find Anyone to Help Yet
- Vision Digital India, A.P. govt. sign MoU to set up AI training centres
- Clark State, Mark Cuban Foundation to host free AI academy
Comments
Please log in to post a comment.