pipette
ESEspañol

Friday, October 9

2,354 new papers appeared. We read every one. These 20 are worth your time.

Each dot is one paper. The pink ones made today's edition. Point at any dot to see its title.

Today's edition

Picked for broad interest, a real step forward and honest claims, with no field allowed to take over.

  1. Prehistoric global migration of vanishing gut microbes with humans

    Our findings indicate that a diverse gut microbiome co-migrated with humans worldwide and has persisted over millennia.
    AI summary

    These groups share 1,231 microbial species, most rare or missing in industrialized populations. Genetic patterns show these microbes co-migrated with humans, splitting when humans left Africa and reached the Americas. The findings indicate this diverse gut microbiome traveled with humans worldwide for millennia. Many of these species are now disappearing from industrialized populations, with uncertain health consequences.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    Peer-reviewed journalBroad interest
  2. A thorium-229 optical nuclear clock with feedback loop

    Here we realize a thorium-229 nuclear clock by stabilizing a continuous-wave laser to the 148-nm nuclear transition with rapid feedback based on absorption spectroscopy 7 .
    AI summary

    The researchers built a nuclear clock using thorium-229 atoms inside a crystal at room temperature. They stabilized a laser to a 148-nanometer nuclear transition using rapid feedback from absorption spectroscopy. The clock reached a fractional frequency instability of 3 × 10⁻¹² per square root of averaging time in seconds. They compared this subharmonic continuously with a ytterbium single-ion clock. The team used the clock to test models of ultralight dark matter by searching for energy fluctuations over 20 seconds to 1 day.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    Peer-reviewed journalClaims a big step
  3. Autonomous thermodynamic cycles via robotic mobility and sensing

    Here, we introduce autonomous thermodynamic cycles enabled by robotic mobility and sensing, allowing robots to perform thermodynamic cycles by accessing spatially varying temperature fields.
    AI summary

    Traditional thermodynamic cycles need fixed hot and cold sources, which limits where they can work. The authors built robots that move through spaces with changing temperatures to run thermodynamic cycles. They used gas-filled capsules that switch between energy states as they travel across temperature gradients. The system harvests and stores energy while optimizing its path to balance movement cost against energy gain. This creates an artificial version of living things that search for energy across their surroundings.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    Preprint
  4. A nuclear clock synchronized to 229Th

    Here we demonstrate a 229 Th nuclear clock by stabilizing a continuous-wave, narrow-linewidth 148.4 nm vacuum-ultraviolet (VUV) laser 18 to a resolved, weakly temperature-sensitive nuclear transition 17,19 in 229 Th:CaF 2 crystals 20–22 .
    AI summary

    The researchers built a nuclear clock using a special laser locked to a nuclear transition in thorium-229 atoms embedded in calcium fluoride crystals. They used a 10-microwatt vacuum-ultraviolet laser source and phototube detection to create a fast, high-quality signal for clock stabilization. The clock achieves a fractional frequency instability of 5 × 10⁻¹³ divided by the square root of the averaging time in seconds. Measurements from two independently made crystals agreed at the 10⁻¹³ level and matched prior results from other crystals. These results show that laser-controlled nuclei can work as practical clock references and offer a reproducible platform for compact nuclear clocks and precision physics tests.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    Peer-reviewed journalClaims a big step
  5. Neanderthal woodworking at Abric Romaní: Insights into the role of wood in a Middle Palaeolithic domestic context

    The Abric Romaní assemblage highlights the importance of wood within Neanderthal technological systems and provides previously unknown insights into the role of organic materials in Middle Palaeolithic societies.
    Peer-reviewed journalClaims a big step
  6. NavGPT-3: Harnessing Context in a Hierarchical Navigation Runtime

    Beneath it, our action policy NavGPT VLA, trained on 19.28M examples, allocates visual tokens using codec allocation, in proportion to scene change; its 8B model alone reaches 74.51 SR on R2R-CE and leads RxR-CE with 78.19 SR. With the complete harness, NavGPT-3 sets the state of the art on R2R-CE (81.51 SR) and, for the first time, brings an autonomous agent to human level: on RxR-CE it matches human followers in success (90.43 vs. 90.4 SR) and path fidelity (78.47 vs. 77.7 nDTW) at 1 min 22 s per episode, versus roughly 3 min for a human.
    AI summary

    NavGPT-3 connects a language model that reasons and plans with an action policy that controls physical movement. An operating-system-like runtime manages both as separate threads, switching control when needed so the robot can react quickly to real-world events. The action policy, called NavGPT VLA, uses an 8 billion parameter model trained on 19.28 million examples. The complete system matches human performance on the RxR-CE navigation task with 90.43% success rate versus 90.4% for humans. The design shortens reaction time from 3-19 seconds per language model decision to 0.5-1 seconds per action policy step.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big step
  7. A solid magnesium zincate for hydrogen-methane evolution

    In this context, we report a low–molecular weight magnesium hydridomethylzincate, Mg[(CH 3 )ZnH 3 ], a crystalline, hydride-rich organometallic compound that does not rely on bulky supporting ligands.
    Peer-reviewed journalClaims a big stepReal-world use
  8. Trillion-scale integrated framework for high-throughput materials databases and seamless sharing

    Here we propose an integrated framework that enables large-scale exploration with reduced resource demands.
    AI summary

    The authors built a framework to explore huge numbers of material combinations without needing massive computing power. Their system uses parallel computing, message queues, containerization, and a lakehouse design with columnar storage. They created a database of ten trillion high-entropy alloys in one week, using 17.5 terabytes. Their method cuts runtime-memory cost by 45%, storage by 90%, and speeds up queries up to 180 times. A full database search finishes in under eight minutes using less than four gigabytes of memory. They screened over 50 billion compositions with experimental validation, confirming practical use for materials discovery.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    Peer-reviewed journalClaims a big step
  9. SDPAD: A Fully Spike-Driven Pipeline for End-to-End Autonomous Driving

    We present SDPAD, a fully spike-driven end-to-end planning pipeline that closes this gap.
    AI summary

    The authors created SDPAD, a fully spike-driven system for end-to-end autonomous driving. SDPAD converts a pre-trained artificial neural network perception stack into integer-spike form. On the nuScenes benchmark, SDPAD achieves 0.40 meter average error and 0.12% collision rate. SDPAD uses 69.9 millijoules, less than 2% of recent artificial neural network baselines. On NAVSIM, SDPAD reaches 86.3 PDMS, surpassing the previous spiking neural network planner by 4.3 points.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big stepReal-world use
  10. Bone regulates locomotion by sustaining motoneuronal mitochondrial function

    This study reveals that through OCN, the skeleton influences motor functions by fine-tuning motoneuronal mitochondria, and suggests that age-related motor decline is, at least in part, reversible.
    AI summary

    Bone releases a hormone called osteocalcin that signals to motor neurons in the spinal cord through a receptor named GPR158. Without GPR158 or osteocalcin, mice have worse movement and lower levels of an enzyme needed for nerve signaling. Osteocalcin signaling helps control how motor neurons recycle their mitochondria and maintain their energy production. Boosting mitochondrial function with drugs or genetic tools can fix movement problems in mice lacking GPR158. Giving osteocalcin to old mice reverses age-related movement decline, but only if motor neurons can perform autophagy and mitophagy.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    Preprint
  11. Skill Constellations: Tracing the Supply Chain of Agent Skills on GitHub

    We contribute the first dated copy network of agent skills, built from the git history of every SKILL.md in GitSkills and covering 2,193,119 skill adoptions across GitHub, together with an interactive viewer.
    AI summary

    Agent skills are instructions and scripts that AI coding agents run with user permissions, and developers share them by copying between repositories. A few repositories are the source of almost all copies, but GitHub stars do not identify these key sources. Skill copies almost never change with their source, so security fixes at the source rarely reach copies. Their model ranks repositories for audit, and reviewing the top 100 prevents 14.9% of later adoptions of high-risk skills.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big stepReal-world use
  12. Programming N-linked glycan composition through de novo protein design

    These results establish steric control of N-glycan processing as a genetically encodable design principle and extend computational protein design to functional glycoproteins with programmed composition.
    AI summary

    N-linked glycans are sugar molecules that attach to proteins and help cells communicate and recognize each other. These simpler sugars still worked biologically: they stopped E. coli bacteria from sticking to human bladder cells and bound to immune proteins. The work shows that sugar processing can be controlled through protein design, opening the door to custom-made glycoproteins.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big stepReal-world use
  13. Stop My Dancing! Understanding, Detecting and Attributing Motion-Aware Deepfake Videos

    Based on the observations obtained from analysis on dataset, we propose MoDA, the first defense framework tailored to detect and attribute MAD videos.
    AI summary

    Pose-guided diffusion models can now create fake videos of moving people, reaching hundreds of millions of viewers. The authors built the first benchmark for these motion-aware deepfakes, with over 1.5 million frames from real and synthetic videos. They found that these fake videos show subtle but reliable signs: high-frequency artifacts and model-specific patterns at motion boundaries. MoDA achieved 94.8% detection accuracy and 91.5% model attribution accuracy, with gains of 10% to 25% over prior methods. It also worked on unseen commercial platforms and internet videos, and held up against adaptive attacks.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big stepReal-world use
  14. Three-dimensional imaging of isolated membrane-protein complexes in vacuo with an X-ray laser

    Here we report a three-dimensional (3D) reconstruction of an isolated 1-MDa membrane-protein complex, photosystem I (PS I), from single-particle diffraction data.
    AI summary

    They reduced background noise by 80% by partially replacing gas with helium. The disc-shaped structure, about 22 nm wide and 10 nm thick, matches the expected size of the protein complex. This shows that membrane-protein complexes can be imaged in a vacuum with X-ray lasers.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big step
  15. Neural Network Verification for Deep Joint Source-Channel Coding

    We present the first bound-propagation framework for verifying DeepJSCC's decoder, bounding worst-case reconstruction error over a given wireless channel's noise region.
    AI summary

    Deep joint source-channel coding uses neural networks to send data wirelessly, but reconstruction quality can drop sharply under attacks and channel noise. The authors created the first method to formally bound worst-case reconstruction error for this system. They adapted existing verification techniques to handle three challenging components: parametric rectified linear activations, transposed convolutions, and Rayleigh fading. They also used Lipschitz-regularized global robustness training, which improved robustness and enabled tight certification for the first time. On image transmission models, their approach lowered median certified bounds by up to 41% and certified about ten times more safe cases than prior methods. Real-world tests on radio hardware confirmed the certificates hold, with worst observed error at 0.082 against a certified bound of 0.128.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big stepReal-world use
  16. Ecology of AI Agents: Collaboration Creates a Population Threshold for Takeoff

    This creates a critical population threshold: below it, the population declines; above it, the population takes off, even though individual-agent capability has not changed.
    AI summary

    AI agents can now carry out real cyberattacks, improve skills as their numbers grow, and work together toward harmful goals. The authors developed a theory about how populations of these agents grow or shrink based on their hacking ability. Without teamwork, a population only grows if each single agent is skilled enough. With teamwork, there is a critical population size: below it, the group dies out; above it, the group explodes in numbers. The authors recommend testing larger groups gradually and measuring how hacking skill scales with group size.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintReal-world use
  17. Psychedelics restructure information in the human brain

    In sum, our results suggest that the psychedelic brain is one in which cortical populations become alike, but not necessarily together.
    AI summary

    The researchers used a mathematical method to study how five psychedelic drugs change information flow in the brain. They analyzed data from 80 people who took LSD, psilocybin, ketamine, DMT, or 5-MeO-DMT. Their findings suggest that under psychedelics, brain regions become more similar to each other but do not necessarily act in unison. The results may help explain how these drugs change conscious experience and treat mental health conditions.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big step
  18. SignRAG: Unified Retrieval-Augmented Gloss-Free Sign Language Translation

    To address this limitation, we propose SignRAG, a unified framework combining hierarchical pretraining, target-domain retrieval augmentation, and retrieval-aware reinforcement fine-tuning.
    AI summary

    The authors created SignRAG, a new system for translating sign language to text without needing gloss labels. The system achieved state-of-the-art results on multiple benchmarks. It is the first gloss-free method to outperform gloss-supervised methods on all metrics for CSL-Daily. The authors have released their code and models on GitHub.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big stepReal-world useCode available
  19. Causal-fate dynamics of unrealized influence

    Here we formulate causal-fate dynamics, in which generated influence may be realized, remain latent, or be transformed by subsequent dynamics, and give an exact finite-transport representation when the relevant maps are specified.
    AI summary

    Many systems create influences whose effects are not fully used up right away. The authors developed causal-fate dynamics to track how such influence can stay hidden or change over time. The worm model suggests but does not prove that unused signals between neurons may affect later activity.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintCode available
  20. Neonatal overfeeding accelerates age-related sarcopenia and metabolic disorders through lifelong gut mucosa and microbiota programming in mice

    Here, we show that neonatal overfeeding acts as a developmental insult that accelerates age-related sarcopenia in both sexes.
    AI summary

    Neonatal overfeeding in mice accelerates age-related muscle loss and shortens lifespan. Early overnutrition disrupts gut goblet cell development, causing lasting mucus barrier defects. These defects reduce Akkermansia muciniphila bacteria, which lowers acetate and impairs muscle mitochondria. Adding A. muciniphila or acetate restores muscle function in aged mice. The findings suggest early nutrition optimization could help prevent muscle aging.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    Peer-reviewed journal