Tuesday, October 20
3,206 new papers appeared. We read every one. These 20 are worth your time.
Today's edition
Picked for broad interest, a real step forward and honest claims, with no field allowed to take over.
Bombadil: A Foundation Model for Plant Electrophysiology
Here we present Bombadil, a foundation model for plant electrophysiology, pretrained without labels on over 4 million hours of electrical signals recorded from 100 plant species in ten distinct environments, including laboratories, greenhouses, commercial orchards, and natural forests.
AI summary
Plants use electrical signals to send information about their health through their bodies. Previous machine learning models could only predict one specific stress type and did not work across different plants or settings. The authors created Bombadil, a foundation model trained on over 4 million hours of signals from 100 plant species in ten environments. Bombadil learned basic patterns of plant electrical signals across many species and settings. The model then tracks seven stress indicators in plants it never saw during training.
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 useDe novo Design of a Vanadium Haloperoxidase
We addressed this challenge by utilizing a machine-learning-guided workflow to design VAN1, a highly stable 20 kDa protein that binds VO43- and brominates activated aromatic substrates, including L-tyrosine.
AI summary
The researchers designed a new protein called VAN1 that binds a vanadate cofactor and adds bromine to certain molecules. They used machine learning to guide the design process. X-ray structures confirmed the designed active site matched the computer model. Mutations improved vanadate binding 15-fold and bromination activity 2.6-fold. VAN1 is the first newly designed enzyme that forms carbon-halogen bonds.
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 stepHow exposed is work to robotics? Occupational, industry, and geographic measures
We construct the first abilities-based exposure measures for robotics and physical AI, and the first to span occupations, industries, and geographies.
AI summary
The authors built the first measures of how robotics and physical AI could affect specific jobs, industries, and locations. They created a taxonomy of 16 robotics capabilities and used a language model to score how each relates to 52 work abilities. Robotics exposure is highest in physically demanding work like construction, farming, transportation, and production. Unlike generative AI, robotics exposure is higher in lower-wage jobs with less education and less creative work. These measures show technical relatedness, not actual adoption, so they represent an upper limit on possible automation.
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 stepScam Town: Can Scam Education Spread Like a Vaccine?
The largest gains came when an official warning, from a bank, a senior centre, or an agency, reached most of a town and residents then talked about it: in simulation (a combination we did not test live) the two together cut scams by about half.
AI summary
The authors built Scam Town, a simulated community where AI plays residents who can warn each other after scam calls. Warnings spread by word of mouth helped but did not go viral: each warning led to fewer than one more. Everyday talk cut scams by roughly a fifth across 80 simulated towns. The largest gains came when official warnings reached most residents and they then talked about it, cutting scams by about half. Because residents were AI stand-ins, the findings are predictions to test with real people, not proof.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintBroad interestEasy to readFlaw-tolerant recycling of waste tires into high-performance rubber
When the flaw-tolerant matrix is incorporated with up to 80 wt% of commercially available waste particles, the resulting composites still maintain strength.
AI summary
About one billion tires reach the end of life each year. Recycled tire rubber is usually downcycled into low-value products, not new tires. The authors designed a flaw-tolerant matrix that prevents waste particles from causing cracks. Their matrix contains two intertwined networks: entangled natural rubber chains and carbon aggregates. Composites with up to 80 wt% waste particles still maintain strength.
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 useUnbounded Losses, Unchecked Inputs: Measuring the Safeguards of LLM Options-Trading Agents
We present the first measurement of the options agents that brokers and developers ship, scored by the safeguards they enforce and by the worst loss their documented order paths can reach.
AI summary
The authors examined 26 systems that let large language models trade options, including all broker interfaces and popular open-source projects. They found that 17 of these systems can execute trades without any required human approval. Thirteen systems, including 6 of the 8 broker interfaces, can cause unlimited financial losses. None of the 26 systems verify where their text instructions come from or double-check market data. The authors conclude that options-trading agents need strict loss limits enforced at the point where orders are placed.
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 useSolving a Giant Rubik's Cube with a Quadruped Robot
We demonstrate this capability by autonomously solving a giant Rubik's Cube with a quadruped using only its feet.
AI summary
Researchers built a system that lets a four-legged robot solve a giant Rubik's Cube using only its feet. The robot uses onboard cameras to estimate the cube's position and colors from partial views while moving around it. Learned skills let the robot flip the cube to reorient it and turn layers by pinning with one foot and pushing with the other. The system completed three autonomous solves on real hardware, each taking about 30 steps over 8-9 minutes. The results suggest that combining onboard perception with foot-based manipulation enables quadrupeds to complete extended multi-step physical tasks.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintHPV-Associated Tonsillar Intraepithelial Neoplasia and Oral High-Risk HPV mRNA as a Candidate Biomarker
These findings establish the first evidence for HPV-associated tonsillar premalignant lesions and suggest oral HR-HPV mRNA as a candidate biomarker.
AI summary
The authors screened 1,180 people having tonsillectomy for benign conditions. They found five with oral high-risk HPV mRNA. Three of these five had HPV-associated intraepithelial lesions in their tonsils. A fourth person with HPV16 infection also had a similar lesion. The findings suggest oral high-risk HPV mRNA as a candidate biomarker.
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 useGenetic determinants of protein phosphorylation in humans
These findings establish protein phosphorylation as a distinct, informative molecular phenotype for GWAS, linking genetic variation, cellular signalling, and human disease risk.
AI summary
The authors conducted the first genome-wide study linking genetic variants to protein phosphorylation levels in humans. They analyzed over 8,000 phosphorylation sites in lung and kidney tissue from 324 participants. They found 46 phosphorylation QTLs that act independently of protein abundance. Some disease-linked genetic variants matched these phosphorylation sites, suggesting phosphorylation affects disease risk through mechanisms missed by protein abundance studies. The authors conclude that phosphorylation provides distinct information for connecting genetic variation to disease.
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 stepRopedia W0: Scaling Loco-Manipulation World Models across Embodiments with Human Experience
We present Ropedia W0, the first egocentric loco-manipulation world foundation model spanning humans and robots with diverse embodiments.
AI summary
The authors built Ropedia W0, the first world model that handles both walking and object handling for humans and robots alike. A shared movement description captures walking via camera positions and hand movements via projected stick figures, with robot figures built from joint calculations and hand-checked points. Tests on five benchmarks and new human videos show top results, from no extra training to full retraining. More human training data improves results for both human and robot tests. After compression, the model runs in real time for minutes, allowing live virtual reality use.
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 availableHuman brain white matter charts in international populations
Here, we established the first set of population-specific WM charts of 198 morphological and microstructural phenotypes for both Chinese (N=19,215) and Western (N=15,584) populations across the human lifespan.
AI summary
The authors created the first brain charts for white matter specific to Chinese and Western populations. They measured 198 structural and microscopic traits across 19,215 Chinese and 15,584 Western participants of all ages. Both groups showed similar overall patterns of brain development and aging, but timing differed by brain pathway and measure. Chinese participants matured earlier in movement and sensory pathways, but later in complex thinking and emotional pathways. Charts matched to the Chinese population better predicted their brain traits and detected more disease-related abnormalities than Western-based charts.
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 useMeasuring the Quantum Phase of Multiphoton Systems
Here, we resolve this problem by measuring, for the first time, the quantum phase of multiparticle systems containing up to fifteen particles.
AI summary
The quantum phase of multiparticle systems has been a mystery because no operator could represent it faithfully. The authors measured this phase for the first time in systems with up to fifteen particles. They used a new tomography method combining homodyne interferometry with photon-number-resolving detection. This allowed them to reconstruct photon-number and phase representations and establish their uncertainty relation. The results may help fundamental quantum physics and scalable multiparticle quantum technologies.
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 stepCross-species transcriptomic atlas of X-linked Myotubular Myopathy identifies unique, treatment responsive myofiber populations
Overall, we address a long-standing question in the field (i.e. the molecular identity of the unusual appearing pathognomonic muscle fibers in XLMTM), provide a framework for understanding transcriptional reprogramming of the XLMTM myonuclear niche, and identify discrete, treatment responsive populations that drive disease pathology.
AI summary
X-linked myotubular myopathy is a congenital muscle disease caused by changes in the MTM1 gene, leading to severe muscle weakness and early death. Treatments largely reversed these unique gene signatures across species.
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 stepMicrocomb spectral fingerprinting enables parts-per-trillion methane detection with low humidity cross-sensitivity
By replacing physically long interaction paths with resonant light recirculation in an ultrahigh-Q microresonator, this architecture brings compact optical sensing into the parts-per-trillion regime while exploiting distributed spectral information for chemical identification.
AI summary
The authors built a tiny glass sensor that traps and recirculates light to detect gases. They detected methane at 178 parts per trillion, with little interference from water vapor up to 90% humidity.
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 usespora: A Unified Multimodal Dataset and Platform for Spatial Proteomics
We present , a large-scale multimodal SP dataset of 13,546 standardized multiplex images from 5,477 patients across 35 cohorts and five major SP technologies.
AI summary
Spatial proteomics maps where proteins sit in tissue at single-cell detail. The authors built spora, a dataset of 13,546 standardized images from 5,477 patients across 35 cohorts. It includes cell maps, clinical records, and paired H&E slides for virtual staining training. Two tools, spora [io] and spora [bench], standardize data handling and test models on twelve cohorts.
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 useSpaceflight-associated Motion Sickness in Commercial Crew Astronauts
These findings establish a systematic baseline for spaceflight-associated motion sickness, informing its operational impact and salience as human spaceflight missions expand in accessibility and scope.
AI summary
This study is the first to measure space motion sickness in commercial astronauts. Eight crew members on two short commercial missions were evaluated. 87.5% developed space motion sickness, and 25% had symptoms for at least three days in orbit. 62.5% of crew members vomited. After returning to Earth, 50% had motion sickness, which was less severe and shorter than space motion sickness.
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 useFortress: Fork Resistant Transparency logs Relying on Social graph Structure
We formalise a new primitive, Social-Graph based Forking-Resilient Transparency (sg-FORT), which captures the security guarantees of a BB-free transparency system and we construct Fortress Fork Resistant Transparency logs Relying on Social graph Structure, to our knowledge the first such protocol.
AI summary
Transparency systems need all honest users to see the same data from a server, or catch cheating. Current designs rely on external tools like bulletin boards or blockchains to ensure this. The authors propose using the social graph of users themselves as a replacement. They create sg-FORT, a new method where either all users agree or someone proves the server cheated. This works even against malicious servers, unless they compromise users who are close together.
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 stepGlobally Optimal Estimation of Nested Logit Models
We develop, to our knowledge, the first globally convergent optimization framework with an explicit global-accuracy guarantee for joint nested-logit estimation.
AI summary
Nested logit models help predict demand, but standard methods cannot guarantee the best possible estimates. Computation time grows polynomially with data size when the number of nests is fixed. The method uses a spatial branch-and-bound algorithm and provides certificates of solution quality. Tests on eight datasets showed the method certifies all solutions and reveals inferior local estimates that other solvers accept.
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 stepStreamAhead: Real-Time Streaming Video Question Answering via Adaptive Precomputation
We present StreamAhead, an algorithm-system co-design that exploits this trade-off through adaptive precomputation: dynamically controlling how much computation to complete for each video segment before a query arrives.
AI summary
StreamAhead is a new system for answering questions about live video streams in real time. It precomputes different amounts of data for different video segments before questions arrive. Each segment stays at a level from simple visual features to complete memory caches. The system schedules work across CPU and GPU to keep video processing on track. Tests show up to 10.4x faster response times than other methods, with 30 frames per second and under 0.5 second response time.
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 useMultiple propagating modes in random polycrystals
Here, using grain-scale finite-element simulations of three-dimensional polycrystals, we resolve the coherent field in the frequency--wavenumber plane and directly observe the predicted multiplicity for both transverse and longitudinal excitations: a single mode at low frequencies gives way to two in the geometric regime as the wavelength approaches the grain correlation length.
AI summary
At low frequencies, one wave mode appears; at higher frequencies, two modes appear as wavelength nears grain size. How much each wave fades determines which modes actually travel and can be detected. This study shows for the first time that coherent waves in random polycrystals are inherently multimodal.
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