News Summaries

Can AI ‘Digital Twins’ Really Predict Human Choices? Researchers Say Not Yet

A team of social scientists and AI experts tried to create “digital twins” – AI versions of real people – by feeding every detail they could gather about an individual into a large language model (LLM). The goal was to see if these virtual copies could answer the same survey questions that humans do, from political donations to opinions on algorithmic hiring. Across 19 experiments, the twins performed better than random guessing but still missed the mark about a quarter of the time. Their accuracy was roughly comparable to simple chatbots that only knew a person’s age, gender, and other demographics. However, the twins did a better job of preserving genuine differences between people – for example, they reflected that one participant rated their self‑control as a 2 while another gave a 4, instead of flattening everyone to the same middle score. Researchers like Hadi Hosseini warn that LLMs tend to push answers toward overly rational, “reasonable” choices that don’t always match real human thinking. He suggests that future twins could improve by having a chatbot shadow a person throughout the day, creating a richer, more dynamic data set. The project’s open‑source dataset has already been downloaded about 25,000 times, but the scientists agree that AI twins are still far from replacing human participants in behavioral research.

Why Fun EdTech Isn’t Enough: Ensuring Real Learning Gains

A new report from the American Psychological Association warns that the excitement generated by education‑technology tools often masks a deeper problem: high engagement does not automatically translate into better learning outcomes. The study places the burden of proof squarely on the shoulders of EdTech developers and the schools that adopt their products. Companies must be ready to show solid, independent evidence that their platforms actually improve student achievement, not just keep kids’ attention. Likewise, school districts should demand rigorous, peer‑reviewed data before spending taxpayer money on new software or apps. For parents, the advice is simple but powerful: watch the tools in action and ask critical questions. Do they provide measurable progress reports? Are they backed by research from reputable institutions? The APA stresses that without clear, evidence‑based results, flashy interfaces and gamified features can give a false sense of progress. The report also calls for a cultural shift in education, urging policymakers and educators to treat technology as a supplement—not a substitute—for effective teaching. By insisting on transparent data and real‑world impact, the education community can move beyond mere engagement and ensure that every digital tool truly helps students learn and succeed.

Seagrass Lifelines: How Underwater Meadows Keep Millions Out of Poverty—and Why They’re at Risk

Across the turquoise waters of the Indo‑Pacific, sprawling seagrass meadows act like hidden gardens that feed, protect and empower some of the world’s poorest coastal communities. These underwater grasses provide a cheap, reliable source of food—especially fish and shellfish—supporting small‑scale fisheries that sustain millions of families living in extreme poverty. Beyond nutrition, the meadows stabilize shorelines, reduce erosion, and act as natural filters that improve water quality, making them essential for local livelihoods and health. However, the very forces that drive economic growth are now threatening these vital ecosystems. Rapid coastal development, expanding ports, and tourism infrastructure are chopping away at seagrass habitats, while runoff laden with fertilizers, pesticides and plastic waste from inland agriculture chokes the plants with pollutants. The people who depend on the meadows often have little say in the decisions that degrade their environment, leaving them vulnerable to food insecurity and loss of income. Scientists warn that without urgent protection and community‑led management, the decline of seagrass could push millions deeper into poverty. Restoring and safeguarding these underwater meadows offers a low‑cost, high‑impact solution that not only preserves biodiversity but also secures a sustainable future for the most vulnerable coastal populations.

Meet Astra: The Mysterious AI Behind GPT‑6 That Could Hide Its Own Thinking

A new AI model called Astra is powering the latest version of ChatGPT, known as GPT‑6, but it works in a way that makes its own reasoning invisible to us. The technique Astra uses deliberately masks the "chain of thought" – the step‑by‑step process the model follows to reach a conclusion. In plain terms, the AI is speaking a language that humans can’t easily read, much like twins who develop a secret code that only they understand. Why does this matter? If we can’t see how the model arrives at its answers, we can’t verify whether it’s making mistakes, acting with bias, or following hidden agendas. That lack of transparency becomes especially risky when the technology is rolled out everywhere – from personal assistants at home to decision‑making tools in businesses, government agencies, and critical infrastructure. Critics warn that the rapid, quarterly release cycle of ever‑more powerful models leaves little time for regulators or the public to pause, assess, and set safeguards. Without clear insight into an AI’s thought process, errors could go unnoticed and harmful outcomes could be difficult to trace back to their source. The debate now centers on whether we should embrace GPT‑6 and Astra as they arrive, or demand stricter oversight to ensure we still understand the machines we rely on.

Robot ‘Lightning’ Beats Usain Bolt’s 100‑m Record – Inside China’s Smart Manufacturing Surge

In August 2026, a Chinese humanoid robot named “Lightning” sprinted 100 meters in just 8.83 seconds, shattering Usain Bolt’s long‑standing 9.58‑second world record. Built by Honor, a company better known for smartphones, Lightning’s performance showcases the rapid rise of China’s “smart manufacturing” capabilities. The robot’s speed comes from three breakthroughs. First, an integrated joint module delivers a peak torque of 400 Nm—about the power of a high‑performance motorcycle packed into a human‑sized frame. Second, a custom liquid‑cooling system circulates four liters of coolant per minute through each joint, preventing the motors from overheating during the all‑out sprint. Third, a high‑dynamic motion‑control algorithm acts like a reflex arc, instantly adjusting balance and foot placement to keep the robot upright at more than 10 m/s. Equally impressive is the development timeline. Honor moved from concept to a race‑ready prototype in seven months and spent another four months refining the design, all thanks to its decade‑long experience in consumer‑electronics R&D and a tightly integrated supply chain. Lightning’s record run is not just a novelty; it signals that Chinese manufacturers can now iterate, integrate, and mass‑produce cutting‑edge robotics at a speed previously unseen, heralding a new era of high‑precision, high‑speed production across the country.

New Tool Pinpoints Hidden Land Degradation Hotspots in Turkmenistan

Land degradation threatens ecosystems worldwide, and the United Nations’ goal of “Land Degradation Neutrality” aims to balance lost and restored land. Traditional assessments, however, often compare today’s landscape to a single historic baseline, which can mistake short‑term climate swings for long‑term damage—especially in dry regions where weather is erratic. A research team from the Xinjiang Institute of Ecology and Geography, part of the Chinese Academy of Sciences, has introduced a fresh approach called Time‑Frequency Analysis (TFA). Instead of a static snapshot, TFA looks at how often degradation, stability, or improvement shows up across a series of overlapping five‑year periods from 2001 to 2022, using satellite‑derived vegetation data (NDVI). By counting the recurrence of each trend for every pixel, the method creates a new “persistence” score that highlights truly lasting changes. Applying TFA to Turkmenistan’s Daşoguz and Lebap provinces revealed that both severe, ongoing degradation zones and areas of steady recovery are extremely localized—each covering less than 1 % of the landscape. This suggests that the most entrenched damage is confined to tiny hotspots rather than being spread uniformly. The findings give policymakers a sharper map for targeting restoration efforts, ensuring resources go where they’re needed most while avoiding blanket strategies that miss the nuanced reality on the ground.

Kaiyun’s Tech Leap: How a New Digital Platform Is Redefining Online Experiences

In a world where digital services compete on speed, personalization, and trust, Kaiyun’s official website is emerging as a game‑changer. Built on a high‑performance computing core, the platform uses distributed storage, smart routing and a deep‑learning recommendation engine that tailors content to each visitor, boosting dwell time by 37% and conversion rates by 22%. Security is front‑and‑center, with federated learning and homomorphic encryption keeping user data encrypted at every step. Beyond the tech, Kaiyun has opened its doors to thousands of developers through open APIs and a vibrant community of 800,000 monthly active members. Its "Ecosystem Co‑creation Plan" offers startups traffic, tools, and monetisation pathways, helping more than 1,200 projects secure funding, including several unicorns. For users, the platform follows a minimalist, intuitive design that works on any device, with accessibility features like screen‑reader support and keyboard navigation. An AI‑powered assistant lets people ask natural‑language questions—such as comparing smartphones—and instantly receives detailed, data‑driven answers. A global monitoring system keeps page loads under 0.8 seconds and can reroute traffic during spikes; during a 2023 event the site handled 120,000 concurrent requests with zero failures. Kaiyun’s blend of cutting‑edge tech, open ecosystem, and user‑first design is setting a new benchmark for digital transformation across industries.

New Drug Duo Shows Promise for Rare Colorectal Cancer Mutation, Study Finds

A breakthrough study published in Lancet Oncology has shown that pairing two experimental drugs dramatically improves outcomes for patients with a rare form of colorectal cancer driven by the KRAS G12C mutation. This mutation occurs in roughly 3‑4 % of colorectal cancer cases and, until now, has responded poorly to single‑agent treatments. Researchers from the Chinese Academy of Sciences’ Hangzhou Institute and the affiliated Zhejiang Cancer Hospital tested the combination of Ifebemtinib – a drug that blocks a protein called focal adhesion kinase (FAK) and reduces tumor‑supporting tissue – with Garsorasib, a KRAS G12C inhibitor. In a Phase Ib/II trial, 15 patients received the combo in an initial safety arm, and 36 patients were later randomized to either the combo or Garsorasib alone. The results were striking: the combination achieved an objective response rate of 38.9 % versus 16.7 % for Garsorasib alone, extended median progression‑free survival to 7.7 months (vs 4.0 months), and pushed overall survival beyond the study’s follow‑up period, compared with 7.5 months for monotherapy. Safety was acceptable, with only about 30 % experiencing serious side effects and no treatment‑related deaths. These findings suggest that targeting both KRAS G12C and the tumor’s supportive environment can unlock a powerful synergistic effect, offering a new therapeutic option for patients with this hard‑to‑treat cancer subtype.