NIH Sparks Uproar as VP Vance Ally Installed to Lead Environmental Health Institute

NIH Director Jayanta “Jay” Bhattacharya abruptly named neuroepidemiologist Kyle Walsh—described as a close friend of Vice President JD Vance—as head of the National Institute of Environmental Health Sciences (NIEHS) on 10 October, notifying staff on 17 October. Walsh, formerly at Duke and a onetime Vance staffer, was appointed without a national search, replacing Richard Woychik, who had reportedly been renewed and will move to Bhattacharya’s office to work on HHS Secretary Robert F. Kennedy Jr.’s “Make America Healthy Again” agenda. The $914 million institute runs toxicology and extramural programs. Former NIEHS director Linda Birnbaum and scientists criticized the process and fear politicization, noting 13 of 27 NIH centers lack permanent leaders. A flap over the NIMH director search—letters dismissing academic members later rescinded—has heightened concerns. (Science)

Apple’s Big M5 Refresh: iPad Pro, MacBook Pro, Vision Pro

Apple quietly rolled out M5-powered updates to the iPad Pro, 14-inch MacBook Pro, and Vision Pro, focusing on performance and AI workloads over redesigns. Built on a 3-nm process, the M5’s 10-core CPU/GPU and 16-core Neural Engine promise up to 4× peak GPU compute versus M4, plus faster memory in 256-GB-and-up configs. The iPad Pro gains 120-Hz external display support; the MacBook Pro targets creators needing sustained performance; and Vision Pro gets better battery life, sharper micro-OLED panels, and a comfier strap. Prices start at $999 (11-inch iPad Pro), $1,599 (MacBook Pro), with Vision Pro still $3,499. Preorders opened ahead of the October 22 launch window, signaling Apple’s incremental but aggressive AI hardware cadence. (WIRED)

Apple’s Second-Gen Vision Pro Lands—With Less Fanfare

Apple has introduced a second-generation Vision Pro with modest hardware refinements and software polishing, signaling a reset of expectations around its mixed-reality ambitions. Reporting highlights a quieter release strategy, improved comfort and optics, and battery tweaks, while suggesting Apple has de-prioritized the platform’s near-term expansion in favor of measured iteration. The move comes as developers continue to probe use cases—from spatial productivity to pro media tools—amid a still-nascent app ecosystem and high price. Apple’s approach mirrors a broader industry pivot: fewer splashy promises, more incremental improvements, and tighter integration with AI-accelerated chips. Whether Vision Pro 2 can broaden adoption beyond early enthusiasts will hinge on compelling apps and clearer day-one value. (Ars Technica)

MorphBot Duo Walks, Drives—Then Flies

Engineers unveiled a two-robot system that morphs between walking, driving, and flying modes, showcasing multi-modal locomotion in a compact, coordinated pair. In tests, each robot reconfigures its chassis and propulsion to traverse varied terrain before taking to the air, then reunites for cooperative tasks. The design illustrates how hybrid mobility could extend mission reach for search-and-rescue, inspection, and planetary exploration, where obstacles and energy budgets vary dramatically. The team details control algorithms that smooth transitions and maintain stability as center-of-mass and actuation change. While endurance and payload remain constraints, the platform hints at a future of adaptable field robots that combine the best of legged agility, wheel efficiency, and aerial access without swapping hardware. (New Atlas)

Self-Organizing Light: A New Path for Optical Computing

USC researchers demonstrated an optical device where light “self-organizes” its route using thermodynamic principles—no digital switches or external controllers required. In the setup, photons explore multiple pathways and naturally settle into efficient configurations, potentially reducing control overhead and energy cost in photonic circuits. The team argues this could simplify interconnects in data centers and on-chip communications while enabling adaptive, fault-tolerant routing. Unlike conventional silicon photonics that rely on precise tuning, the approach embraces physical dynamics to find solutions, echoing emerging trends in analog and neuromorphic compute. Still early, the work will need scalability, reproducibility, and CMOS-friendly fabrication to matter commercially—but it points to tantalizing gains for bandwidth-hungry AI and telecom workloads. (SciTechDaily)

Sodium-Ion Battery Leap by… Keeping Water In

A University of Surrey–led team reports a counterintuitive boost for sodium-ion batteries: retain water in a key component instead of removing it. The hydrated structure improves ion transport and stability, pushing performance closer to lithium-ion while using cheaper, widely available sodium. Because sodium-ion chemistries can rely on abundant materials and potentially safer electrolytes, they’re contenders for grid storage and low-cost mobility—especially where lithium supply or price volatility is a concern. The study also notes desalination synergies, with implications for coupling energy storage to water treatment. Commercialization will require cycle-life validation, manufacturing pathways, and compatibility with existing pack architectures, but the tweak suggests overlooked variables in electrode processing can unlock sizable gains. (Phys.org)

Smarter Hydrogen from Ammonia: Control-Oriented Model Debuts

Researchers unveiled a control-oriented model to improve cracking ammonia into hydrogen more efficiently—key for using ammonia as a transportable H₂ carrier. The framework simplifies reactor behavior into real-time equations suitable for embedded controllers, allowing adaptive responses to fluctuating loads and temperatures. That’s vital for mobile or distributed systems that can’t rely on steady-state optimization. Early results indicate higher conversion efficiency and better stability, which could lower energy costs and emissions for hydrogen supply chains. The team’s next steps include hardware-in-the-loop testing and integration with sensors for fault detection. If validated at scale, the method could accelerate ammonia-to-hydrogen systems for fueling stations, maritime applications, and microgrids. (Phys.org)

AI Re-Maps the Mouse Brain in Astonishing Detail

Using spatial transcriptomics and deep learning, scientists produced a mouse-brain atlas with hundreds of newly defined subregions, revealing molecular borders invisible to classical anatomy. A neural network integrated gene-expression maps with structural data, refining boundaries across cortex, hippocampus, thalamus, and more. The result offers a higher-resolution blueprint for neuroscience, enabling targeted experiments, disease modeling, and drug discovery with unprecedented precision. The team emphasizes open data and tools to spur replication and extension to other species. While primarily a basic-science advance, its informatics pipeline—fusing high-dimensional ‘omics with imaging—highlights how AI is transforming biospatial mapping, a pattern likely to inform human brain charts and clinical translation over time. (The Scientist)

Scientists Should Leverage—Not Compete With—AI

A new commentary argues researchers must integrate AI into lab workflows rather than treat it as a rival, especially as regulators tighten expectations around software-as-a-medical-device. The authors point to the FDA’s 2025 draft guidance encouraging “predetermined change control plans,” effectively requiring machine-readable logs of iterative model and firmware updates. Without systematic digital documentation, compliance could become unmanageable. The piece outlines pragmatic steps: provenance tracking, automated validation, and human-in-the-loop checkpoints to preserve scientific accountability. It also cautions against hype, urging labs to target tasks where AI demonstrably improves reproducibility and throughput. The broader message: AI is becoming infrastructure—embrace it with guardrails or risk being left behind as standards solidify. (The Scientist)

When Cells Self-Destruct: New Insight Into Antiviral Triggers

Researchers detail how specialized immune proteins assemble crystalline platforms that activate caspases, driving infected cells to self-destruct and halt viral spread. The work clarifies why the same pathways sometimes cause damaging inflammation, offering targets for tuning responses in severe infections or autoimmune disease. While a biology story at heart, the underlying structural tools—cryo-EM, computational modeling, and protein design—reflect the tech-driven fusion of imaging and AI that’s reshaping immunology. Potential spinoffs include antivirals that nudge the death-fold machinery and diagnostics that read out self-assembly states. The article places the findings in the larger debate over modulating programmed cell death as a therapeutic lever. (Scientific American)

Is AI’s Scaling Obsession Hitting a Wall?

A WIRED analysis surveys mounting pressures on AI’s “just scale it” doctrine—from power constraints and chip bottlenecks to ballooning training budgets and uncertain returns. It spotlights efforts by Modular (founded by ex-Apple luminary Chris Lattner) to optimize how software drives heterogeneous GPUs, alongside concerns that model-size gains are yielding smaller utility jumps. The piece canvasses investors, researchers, and operators wrestling with data center siting, GPU supply, and tooling fragmentation. The near-term outlook: more emphasis on efficient runtimes, domain-tuned models, and better orchestration rather than brute-force parameter races. Whether the industry rebalances in time may decide who profits in the next AI wave. (WIRED)

Fossil Brood Balls Push Carrion-Eating Dung Beetles Back 38 Million Years

A new study overturns the idea that South American dung beetles only recently evolved to eat carrion. Analyzing 5,340 fossilized brood balls from nine formations in Argentina, Chile, Uruguay, and Ecuador, researchers identified “perched” larval chambers—architectural cues of necrophagy—dating to 37.7 million years ago in Patagonia. That predates megafaunal extinctions by tens of millions of years and contradicts the long-standing view that beetles switched diets 130,000–12,000 years ago as dung declined. The findings align with 2020 genomic evidence placing necrophagy’s origin 35–40 million years ago in South America. The team argues open grasslands that arose ~45 million years ago supported abundant herbivores and dung beetles; intense competition likely drove some species to exploit carcasses. Brood-ball architecture thus preserves behavior, revealing beetles as ancient ecological opportunists. (Science)

IMAGE CREDIT: DonkeyHotey