AI is moving into more specialized work, from crop breeding to drug research and software operations. A review of “Breeding 5.0” says AI, plant synthetic biology and repeated design-build-test-learn cycles could shorten development of some crop traits from eight to ten years to four to five. AI can analyze images and sensor data, while CRISPR tools enable targeted edits; maize examples include multiplex editing of EPSPS, ALS and ACCase and genes affecting kernel rows and cob diameter.
A University of Alberta review finds free AI tools can support many routine drug-development tasks, including discovery, preclinical testing and safety monitoring. Open tools named include OpenFold, Chai-1, Protenix and Boltz-2, though the authors say Boltz-2 claims need independent testing. Commercial platforms still dominate full discovery programs and regulatory workflows, where validation, vendor support and legal protections matter. The review also warns that online services may send molecular or biological data to third parties.
At Meta, engineers are testing AI teammates that review alerts, tickets, deployments and incident history, while leaving action approval to people. In a 12-alert overnight test, the teammate prepared a hypothesis and fix that an engineer verified at 2:01, versus 2:39 with a question-answering assistant. In a longer trial, three of 23 items needed human attention, 18 went through normal review and CI, and two remained in progress. Meta says workload reduction looks promising but will not estimate time saved before a rigorous study; high-criticality work remains human-owned. The company says AI now writes about three quarters of its new code.
Another software effort tackles unreliable claims by AI coding agents. Stephen and Cassandra Chin described using Relate and the Archie Rust graph database to assemble evidence packets without an LLM, helping reviewers check claims such as whether tests actually passed. PR Bench scores for Tracetune plus Relate rose from 74 to 76 and 82 to 86; the article does not explain what the score ranges measure. It also describes Strata, a one-click local setup for Windows and Linux to run Qwen3.8-Flash-Next over localhost.
Google.org funds the AI Research Foundations programme, which aims to bring advanced training to 30 universities in Ghana, Kenya, Nigeria and South Africa and reach 30,000 students. Its South African programme launched on November 9 at Stellenbosch University. Free Google Skills courses cover deep learning, neural networks, generative language models, transformers and responsible AI. A five-day workshop through November 13 brings together lecturers and researchers from five South African universities, with a Train the Trainer model and continued support for teaching and student projects.
Researchers Jennifer J. Chen and Hui Li call for child-focused AI standards after reviewing five studies on children ages 3 to 8. Smart toys, digital assistants and adaptive platforms such as Amira can personalize learning, they say, but raise privacy, safety and well-being concerns. They recommend standards developed with developers, policymakers, schools, educators and families, and point to the POWER principles: Purposeful, Optimal, Wise, Ethical and Responsible. In Michigan, more than 90 educators attended Kent ISD’s fourth annual AI Summit in early November to discuss student success and fair classroom access. Keynote speaker Danielle Boyer warned that AI can misrepresent Indigenous languages and cultures without community consent or accurate technology; educators also shared ways to set clear learning goals and model responsible use.
Amazon-owned Ring announced two mobile security products. The $249.99 Ring Drone Cam follows scheduled routes, flies on demand or responds to a Ring camera, sensor or alarm, and streams 2K video with a 148-degree field of view. Ring says it will arrive in the United States on January 27, though it has not received FCC authorization. The $999.99 Ring Rover Cam, a wheeled patrol robot for businesses, is already available. Ring says the Rover’s camera retracts and microphone switches off in sleep mode, while the drone’s camera stays covered and off when docked.
A separate report explains how AI training data is assembled from sources including Common Crawl, reference sites, public code, books, licensed material and human-written or labeled content. Crawlers follow links, with teams managing visit frequency and JavaScript-heavy pages; data is then extracted, language-identified, filtered, deduplicated and screened for private or harmful material. Public web content may still be subject to copyright, usage terms and privacy rules. Public archives can miss recent events, regions and specialized material, prompting targeted crawls.
Practis appointed Ryan Felton chief technology officer on November 11, 2026. He will lead engineering, AI and product strategy, including Voice AI Labs, as the company develops tools connecting learning, practice and feedback for customer conversations. Felton previously held senior software engineering and technology leadership roles at UnitedHealth Group, BLiNQ Media and Pulse Platform. Practis combines AI Coach, AI Customer simulations and analysis of live conversations to support sales-team development and help leaders track progress.
Key Takeaways
- “Breeding 5.0” combines AI, synthetic biology and CRISPR; the review says some crop-trait development could fall from eight to ten years to four to five.
- A University of Alberta review says free AI tools cover many routine drug-development tasks, while commercial platforms remain prominent in full discovery and regulatory workflows; online data-sharing and validation are key concerns.
- Meta’s AI on-call teammate reviewed incidents and produced a fix for engineer approval; in a 12-alert test, verification took until 2:01 versus 2:39 with a question-answering assistant. Meta says rigorous measurement is still pending.
- Relate and the Archie graph database create evidence packets to check AI coding-agent claims without using an LLM; reported PR Bench scores rose from 74 to 76 and 82 to 86.
- Google.org-backed AIRF aims to train 30,000 students across 30 universities in Ghana, Kenya, Nigeria and South Africa; its South African launch was November 9 at Stellenbosch University.
- Researchers call for child-specific AI safeguards for ages 3 to 8, while Michigan educators discussed learning goals, fair access and risks to Indigenous language and cultural representation.
- Amazon-owned Ring announced a $249.99 Drone Cam, planned for U.S. arrival January 27 pending FCC authorization, and a $999.99 Rover Cam that is already available.
- AI training datasets draw on web crawls, books, code, licensed sources and human contributions, but collection and use face quality, privacy, copyright and coverage limits.
- Practis named Ryan Felton CTO on November 11, 2026, tasking him with engineering, AI and product strategy and leadership of Voice AI Labs.
AI and Synthetic Biology Promise Faster Crop Improvement
A review says combining artificial intelligence with plant synthetic biology could speed up crop improvement and help address limits in traditional breeding, including long growing cycles, declining genetic diversity, and climate change. The approach, called Breeding 5.0, uses repeated design, build, test, and learn cycles that bring together genetic, plant trait, and environmental data. AI can analyze images and sensor data to track crop traits, while machine learning can improve predictions for breeding. CRISPR tools, including prime editors and the ePE5max architecture, can make targeted genetic changes; in maize, multiplex editing has been used to target EPSPS, ALS, and ACCase, as well as genes that affect kernel row number and cob diameter. The article says this could cut development time for some traits from about eight to ten years to four to five years.
Review Finds Free AI Tools Cover Many Drug Development Tasks
A review by Neal M. Davies and Raimar Loebenberg of the University of Alberta examines open and commercial AI tools used across drug development. Published in the Journal of Pharmacy and Pharmaceutical Sciences, it covers discovery, preclinical testing, clinical trials, regulatory review, and post-market safety. The authors say freely available tools can handle many routine research tasks, but online services may send molecular or biological data to third parties. Commercial platforms remain dominant for full drug discovery programs and workflows used in formal regulatory filings, where validated software, vendor support, and legal protections matter. The review also discusses open tools such as OpenFold, Chai-1, Protenix, and Boltz-2, while noting that claims about Boltz-2 need independent testing.
Meta Tests AI Teammates to Help On Call Engineers Handle Alerts
Meta says about three quarters of its new code is now written with AI, but operating systems has not become faster in the same way. Chinmay Gandhi and incident response engineers Swaapnika Motapothula and Vlad Tsvang described an AI teammate for each on-call rotation that reviews alerts, tickets, deployments, and past incidents, while leaving engineers to approve actions. In a test involving 12 alerts at 2 a.m., the AI teammate grouped related issues, linked a configuration change and past incidents, and prepared a hypothesis and fix for review; the engineer verified it at 2:01, compared with 2:39 using a question-answering assistant. In a longer trial, 23 items arrived, with three requiring human attention, 18 handled through normal review and CI, and two still in progress. Meta said early signs of reduced workload look promising, but it will not estimate time saved until it completes a rigorous study, and high-criticality work remains human-owned.
A New Evidence Tool Checks AI Generated Code Test Claims
Stephen Chin and Cassandra Chin said AI coding agents can produce inaccurate session logs as they work. They described a pull request that claimed tests passed, even though traces showed bun test and bun x tsc running at the same time, when the commands needed to run one after the other. Their fix uses Relate, a Rust embedded graph database called Archie, to build an evidence packet without using an LLM. Chin said the evidence lets a cheaper Haiku model review claims while a Sonnet model writes. PR Bench scores for Tracetune plus Relate rose from 74 to 76 and from 82 to 86, though the article does not identify what those score ranges measure. The article also mentions Strata, which offers a one-click local setup for Windows and Linux to run Qwen3.8-Flash-Next over localhost.
AIRF Brings Advanced AI Training to African Universities
The AI Research Foundations programme aims to bring advanced AI education to 30 universities in Ghana, Kenya, Nigeria, and South Africa, with a goal of training 30,000 students. Google.org funds the initiative, which launched its South African programme on 9 November at Stellenbosch University’s Biomedical Research Institute on the Tygerberg campus. Its courses cover deep learning, neural networks, generative language models, transformers, and responsible AI, and are free through Google Skills. A five-day workshop running until 13 November brings together lecturers and researchers from Stellenbosch University, the University of Venda, the University of Mpumalanga, the University of the Free State, and the University of the Western Cape. AIRF uses a Train the Trainer model, with ongoing support to help participants teach the curriculum and guide student projects.
Researchers call for child focused AI rules in early education
Jennifer J. Chen and Hui Li reviewed five studies in the first issue of AI, Brain, and Child, focusing on children ages 3 to 8. They say AI tools, including smart toys, digital assistants, and adaptive learning platforms such as Amira, can support personalized learning but also create ethical risks. The authors say early childhood needs its own standards for privacy, safety, and child well being, involving developers, policymakers, schools, educators, and families. They point to the POWER principles, Purposeful, Optimal, Wise, Ethical, and Responsible, as a guide for AI use and oversight.
Ring announces home security drone and business patrol robot
Amazon owned Ring announced the Ring Drone Cam for homes and the Ring Rover Cam for businesses. The $249.99 drone can follow scheduled routes, fly on demand, or respond to a triggered Ring camera, sensor, or alarm, and streams 2K video with a 148 degree field of view. Ring says it will arrive in the United States on January 27, though it has not yet received FCC authorization. The $999.99 Rover is a wheeled patrol robot that is already available, and Ring says its camera retracts and microphone shuts off in sleep mode. Ring also says the drone camera stays covered and off while docked, amid privacy concerns about cameras that move around on their own.
Michigan AI summit explores student success and fair access
More than 90 educators from Michigan and other states attended Kent ISD’s fourth annual AI Summit in early November. The event focused on using AI and technology to support learning, student voices, and fair access to classroom resources. Keynote speaker Danielle Boyer discussed data sovereignty and warned that AI can misrepresent Indigenous languages and cultures when communities do not give consent or the technology is inaccurate. Educators including Sarah Wood and Aaron Romoslawski also shared ways to use AI with clear learning goals and to model good practices for students.
How AI training data is gathered and cleaned from the web
Large language models use text from sources such as Common Crawl, reference sites, public code, books, licensed data, and material written or labeled by people. Teams that collect web pages use crawlers to follow links, while managing how often they visit sites and how they handle pages that need JavaScript. Data teams then extract useful text, identify languages, filter low quality material, remove duplicates, and screen for private or harmful content. Public web pages may still be covered by copyright, usage terms, and privacy rules. Public archives can also miss recent events, some regions, and specialized material, leading some teams to run targeted crawls.
Practis appoints Ryan Felton CTO to lead Voice AI development
Practis named Ryan Felton its chief technology officer on November 11, 2026, as the company expands its AI sales training and Voice AI platform. Felton will guide engineering, AI, and product strategy, lead Voice AI Labs, and connect learning, practice, and feedback for real customer conversations. His past roles include senior director of software engineering at UnitedHealth Group and technology leadership at BLiNQ Media and Pulse Platform, both acquired by other companies. Practis combines AI Coach, AI Customer simulations, and analysis of live customer conversations to help sales teams build skills and help leaders track development.
Sources
- Artificial Intelligence and Synthetic Biology Converge to Accelerate Crop Improvement
- Review Evaluates Open and Closed Artificial Intelligence Across Pharmaceutical Drug Development
- Meta wants on-call to approve, not operate
- Stop trusting PRs that claim tests passed
- African universities gear up for advanced AI training as AIRF programme launches
- Researchers Establish Ethical Frameworks for Artificial Intelligence in Early Childhood Education
- Ring unveils AI-powered security drone that patrols your home and a robot that guards businesses
- AI Summit explores technology’s role in student success, equity
- Where AI training data comes from: Collecting public web data at scale
- Practis Names Ryan Felton CTO to Advance AI Sales Training and Voice AI Innovation
Comments
Please log in to post a comment.