PAPER PLAINE

Researchers created a new way to fit VARMA models—statistical tools that forecast how multiple variables change together over time—that works at practical scale for the first time. By reformulating the math and using Fourier techniques, they cut the computational cost per iteration from growing with data length to staying fixed, allowing the method to handle 10–40 variables where it previously failed or produced unreliable forecasts.

Physicists tested whether fundamental speed-accuracy tradeoffs hold up in quantum systems when interactions are added. The bounds remained valid in some cases but broke down in others depending on how the system's interactions were defined—suggesting that nonlinearities pose a real threat to these limits that theorists had thought were rock-solid.

Language models often get confused by misleading external information—they either blindly trust it and produce wrong answers, or ignore all context and become useless. Researchers created a training method called SCOPE that teaches models to selectively trust context: accept it when it's correct or irrelevant, but reject it when it conflicts with what they already know correctly. The method reduced errors from misleading signals across multiple popular AI models while keeping their accuracy intact when the context was actually helpful.

California's proposed high-speed rail system could reduce flight delays at major airports by drawing passengers away from short-haul flights, saving airlines $51–88 million annually by 2029 and $235–392 million by 2033. The researchers traced how fewer planes taking off from San Francisco, Los Angeles, and San Diego would ease congestion at airports nationwide, reducing costly delays for all departing flights.

When solar and wind replace traditional power plants, grids lose the natural stability those plants provided. Researchers tested whether planning for stability from the start—rather than bolting it on afterward—saves money and picks better solutions. An integrated approach reduced total system costs and favored battery storage systems that can do double duty: storing energy and stabilizing the grid.

Researchers tested whether artificial intelligence methods that enhance low-resolution brain MRI scans preserve small white-matter lesions—signs of stroke risk and neurodegeneration. The dominant problem was erasure: AI tended to delete real small lesions rather than invent fake ones, and this worsened with thicker original scans. One method, ECLARE, recovered small lesions better than others, but none performed as well as the original high-resolution images.

Researchers tested whether ChatGPT could predict how loan returns would change if interest rates were set differently—a type of "what-if" analysis critical for lenders deciding pricing strategy. With better prompting, ChatGPT's predictions improved significantly, reaching performance close to traditional machine-learning models, and the system showed it could reason through cause-and-effect relationships in loan outcomes.

When traffic cameras try to pinpoint a vehicle's location, simply using the center of the detected box produces large errors—especially for distant or elevated cameras. Researchers built a system that predicts where all four corners of a car's base touch the road, then calculates the true center from that shape. On real-world footage, this cut localization errors by over half, dropping median ground-plane error from 5.52 meters to 0.90 meters for medium-range vehicles.

Researchers created a framework that uses AI judges to evaluate whether benchmarks used to test conversational agents are actually good measures of performance. The framework checks three key qualities—consistency, complexity, and how thoroughly benchmarks cover different behaviors—and identifies specific weaknesses. When tested against human judgment and against benchmarks intentionally made worse, the system reliably distinguished between high-quality and low-quality benchmarks.

Most drugs fail in testing because their target wasn't actually causing the disease. Researchers built a knowledge graph that tracks how biomedical evidence changed over time, then used it to predict which drug candidates would advance from Phase II to Phase III trials using only the evidence available when those decisions were made. The approach outperformed direct evidence alone, especially for the 73% of candidates with no direct proof their target mattered.

AI agents that sign documents and authenticate systems currently store their cryptographic keys where any compromised program can steal them—a vulnerability exploited in a real incident in under five minutes. Researchers replaced software-based keys with hardware keystores (like TPMs or smart cards) that perform signing operations inside locked hardware, sending only results back to the AI. Testing against 12 injection attacks reduced success rates from 19.3% to 0%, with no false alarms on legitimate tasks.

A new system called TYTAN can automatically figure out the structure and meaning of data in business databases by combining pattern-matching with AI language models, asking targeted clarifying questions when needed. In tests across eight real databases, it achieved perfect accuracy in finding every entity and executing data retrieval instructions correctly, and correctly identified semantic roles 92–100% of the time.