Reading up on research

China has rapidly evolved from a space newcomer to the United States’ primary rival for dominance across the solar system.

Cloudflare Research introduces Meerkat, a global consensus service using the QuePaxa algorithm to provide strong consistency and high availability for control‑plane data across its 330+ data centers.

A new study finds that deep-sea life benefits from nutrient-rich castoffs of melting icebergs, offering a rare silver lining amid Arctic ice loss.

Using X-ray scanning and machine learning, the Vesuvius Challenge team virtually unwrapped and read the first complete Herculaneum scroll (PHerc. 1667) without physically opening it.

The U.S. government is investing $150 million in technologies that give buildings immune-like systems to fight airborne diseases, following the pandemic.

Nvidia's ENPIRE harness lets AI coding agents autonomously train robots to perform physical tasks like cutting zip ties and installing GPUs.

Robin Hanson argues that large language models could drive progress by systematically finding and resolving contradictions between unrelated fields of knowledge.

Count Anything introduces a dual-granularity point-based model that unifies text-guided object counting across six visual domains, outperforming existing open-world methods.

A multi-year scientific initiative aims to engineer identical natural grass surfaces for all World Cup venues to ensure consistent playing conditions.

New research demonstrates that quantum physics can generate truly random numbers, overcoming the limitations of conventional computers in encryption systems.

An AI system called Gauss completed a mathematical proof formalization in five days, outpacing a human team that had worked for two years, raising concerns about the future role of mathematicians.

Modern LLMs use stacked transformer blocks with attention, feed-forward networks, and residual connections to predict the next token from tokenized input.

This paper proposes a 'Sleep' paradigm for LLMs, combining knowledge seeding via distillation and dreaming via RL-generated curricula to enable continual learning and memory consolidation.

A replication study finds that machine learning cannot identify outperforming mutual funds, but can help avoid underperformers.

A study of Australian early adopters of the 100:80:100 four-day workweek model reports positive impacts on productivity and employee well-being.

A personal guide lists and reviews nine Lean 4 books, offering subjective opinions and suggested learning paths.

Scientists have retired the RCP8.5 climate scenario, raising questions about whether some global warming risks were overstated.

The UK's AI Security Institute, staffed by alumni from OpenAI and Google, is becoming a model for nations assessing emerging risks from advanced AI systems.

SpaceX's Starship V3 introduces major engine and design upgrades, with a suborbital flight planned to test payload deployment and heatshield inspection.

Social media use intensity positively predicts academic procrastination among college students, mediated by psychological capital and moderated by time management disposition.

An OpenAI model has disproved a long-standing central conjecture in discrete geometry, sparking both excitement and debate in the mathematics community.

CARA 2.0 is a sub-$1000 quadruped robot built by Aaed Musa using cheap drone motors and controllers to make dynamic legged robots accessible to hobbyists and researchers.

AlphaEvolve, a Gemini-powered coding agent, has optimized algorithms across genomics, quantum physics, and infrastructure, achieving significant improvements.

Global health officials race to track a hantavirus outbreak on a cruise ship but predict limited spread, as South African analysis finds no viral mutations.

SoundOff uses passive, battery-free 3D-printed metal tags that emit unique ultrasound chirps when moved, enabling private, zero-infrastructure smart home sensing.

Thomas Harriot invented binary numeration around 1601-1605 based on his idiosyncratic method of recording part-ounce weights in power-of-2 ratios, decades before Leibniz.

GLP-1 drugs silence the constant mental chatter about food, prompting obesity researchers to study this phenomenon they previously overlooked.

The Artemis II crew splashed down off San Diego after a 10-day lunar flyby, marking the first crewed moon mission since Apollo.

Since mid-2025, AI models have begun proving new mathematical results at an accelerating pace, with mathematicians reporting discoveries that would have taken weeks now achieved in days.

NASA Artemis II astronauts returned to Houston after a 10-day moon mission, marking an emotional reunion with family and friends.

The system card for Claude Mythos Preview details model capabilities, safety testing protocols, and known limitations for the new AI system.

An astronaut's medical emergency in January left him unable to speak and forced an evacuation of the International Space Station, but the condition remains a mystery.

A study finds that popular AI models give biased, overly agreeable feedback on social situations, undermining their impartiality.

Serial cloning of mice for 58 generations accumulated mutations and chromosomal abnormalities, causing efficiency to drop to near zero, yet surviving clones had normal lifespans.

The article reviews evidence that collagen supplements may improve skin elasticity and joint health, but rigorous studies are limited and benefits likely modest.

The Turing Award has been awarded to Charles Bennett and Gilles Brassard for their 1980s invention of quantum cryptography, an unbreakable encryption method.

Cortical Labs demonstrated its CL1 biological computing platform by having living human brain cells play DOOM, sparking ethical and technical debates.

Donald Knuth publishes a paper showing Claude helped him find and prove a solution to a combinatorial problem, though even-case solutions remain open.

Research indicates pet ownership is linked to lower heart disease risk and longer life, but benefits depend on lifestyle and pet type.

New study suggests super-agers have a unique brain characteristic linked to their superior memory.

Bell Labs pioneered cellphones, satellites, video calls, and foundational AI research through its 20th-century innovation labs.

US science funding cuts and political hostility are driving young researchers abroad, threatening America's dominance in biomedical research.

Trump administration cuts to NIH funding and grants are driving young US scientists to seek positions abroad, threatening the country's biomedical research dominance.

Training sequence model agents against diverse co-players induces in-context best-response strategies that naturally lead to cooperative behavior without hardcoded assumptions or explicit timescale separation.

Energy drinks contain vitamins and plant extracts, but evidence suggests these extra ingredients offer little to no benefit beyond caffeine's stimulant effects.

The backlash over Dr. Peter Attia's ties to Jeffrey Epstein has widened into a debate about his credentials and whether patients can trust longevity medicine.

A study of dogs along the Ukraine front line found surprising behavioral and physiological changes in former pets, researchers report.

Wooden tools dated to 430,000 years ago, discovered in England, are the oldest known wooden tools ever found.

Google releases Project Genie, an AI-powered prototype that lets subscribers create and explore interactive 3D worlds from text or images.

Recent advances in expansion microscopy, protein barcoding, and AI-based connectomics make full human brain emulation plausible within decades, but require massive data acquisition and compute.

The 2025 Web Almanac analyzes 17.2 million websites using HTTP Archive data to understand web trends, accessibility, performance, and security.

A benchmark using the game "So Long Sucker" finds that simpler LLMs can outperform in complex scenarios and that models adjust their honesty based on opponent strength.

Researchers are using Claude as a collaborative AI agent to automate experimental design, data interpretation, and hypothesis generation, compressing months of work into hours.

OpenAI's GPT-5 accelerates research in math, biology, and chemistry, but experts debate whether it can generate truly novel ideas independently.

Citation-aware rubric rewards (CaRR) and C-GRPO training improve deep search agents' reasoning quality and factual grounding over standard binary outcome rewards.

Vaccine development has progressed from serendipitous discoveries to a systematic, atomic-precision science, enabling the design of new vaccines in weeks rather than centuries.

Google's Willow quantum chip solved a benchmark in minutes that would take a classical computer 10 septillion years, demonstrating error correction and advancing quantum computing.

Blocking the 15-PGDH protein via injection regrows knee cartilage and prevents arthritis in aging and injured mice, with potential for human therapy.

A study finds Supreme Court Republican appointees voted for wealthier side 70% of the time in 2022, up from 45% in 1953.

Vaccines provide unexpected off-target benefits for older people, including a reduced risk of dementia, according to multiple studies.

NASA’s Goddard Space Flight Center will close its research library on Friday, discarding or warehousing unique historical documents from early space exploration.

The Fifth Edition of the Red Book collects opinionated readings on database research, from classic RDBMS to weak isolation and large-scale dataflow engines.

A new study suggests that people who reach the pinnacle of their fields typically started off dabbling in multiple disciplines in their youth.

OpenAI researcher John Schulman discusses reinforcement learning scaling, LLM usefulness milestones, and lessons for building AI research teams.

Studies found that a brief conversation with a trained chatbot was about four times as persuasive as a traditional political TV ad in shifting political opinions.

A clinical trial of Eli Lilly's retatrutide, which targets three hormones, resulted in significantly greater weight loss than any currently approved drug.

A Stanford experiment shows its AI hacking bot Artemis can find and exploit vulnerabilities in a real network, sometimes outperforming professional human penetration testers.

A study of 1,250 professionals finds AI used like a junior intern—handling grunt work and first drafts, but never trusted without human verification and editing.

Anthropic's new Interviewer tool conducted 1,250 interviews, finding professionals are optimistic about AI but concerned about job displacement and creative identity.

A researcher extracted a 'soul document' from Claude 4.5 Opus's weights that contains the character training guidelines Anthropic used to shape the model's values and behavior.

LATE dementia, though less severe than Alzheimer’s alone, worsens Alzheimer’s symptoms when the two coexist, according to scientists.

In November 2024, the quantity of AI-generated articles published online surpassed human-written articles, though the proportion has plateaued since May 2024.

Interruption rate, recovery time, and minimum focus block size mathematically determine whether a workday yields deep work, with simulations showing small parameter changes drastically shift productivity.

Gene therapy using a modified HIV virus has effectively cured 48 of 50 infants born with severe combined immunodeficiency, restoring normal immune function.

Pet cats in a small study meowed more at male owners than female caregivers, suggesting felines tailor vocal greetings by owner gender.

MIT研究用Iceberg Index模拟发现AI已可替代美国11.7%的劳动力，涉及1.2万亿美元工资。

Occasional binge drinking—four or more drinks in a sitting—still carries significant health risks, including cancer and heart disease, experts say.

A study of brain scans from nearly 4,000 people identifies five developmental epochs with key turning points around ages nine, 32, 66, and 83.

A meta-analysis of 230 studies finds that adjusting or abandoning unattainable goals reduces stress, anxiety, and depression more than persistent grinding.

A German team achieved the first quantum teleportation between photons from two different quantum dots, a key step toward practical quantum repeaters for secure long-distance communication.

Rapid GPU upgrade cycles threaten to make hyperscalers' hardware unprofitable, creating a financial crisis from accelerated depreciation and risky loans.

Google launched Gemini 3, an AI model that sets new records in reasoning and multimodality, available immediately in the Gemini app, AI Studio, and Vertex AI.

Reinforcement learning provides far fewer bits per FLOP than pretraining until models achieve high pass rates, limiting RLVR's ability to learn new capabilities.