Lifeboat Foundation: Safeguarding Humanity
Toggle light / dark theme

Blog

Category: transportation

Novel alloy withstands extreme conditions, could replace metals used in aircraft engines and gas turbines

A new material might contribute to a reduction of the fossil fuels consumed by aircraft engines and gas turbines in the future. A research team from Karlsruhe Institute of Technology (KIT) has developed a refractory metal-based alloy with properties unparalleled to date.

The novel combination of chromium, molybdenum, and silicon is ductile at . With its of about 2,000°C, it remains stable even at high temperatures and is at the same time oxidation resistant. These results are published in Nature.

High-temperature-resistant metallic materials are required for , , X-ray units, and many other technical applications. Refractory metals such as tungsten, molybdenum, and chromium, whose melting points are around or higher than 2,000°C, can be most resistant to high temperatures.

AI-based model can help traffic engineers predict future sites of possible crashes

In a significant step toward improving road safety, Johns Hopkins University researchers have developed an AI-based tool that can identify the risk factors contributing to car crashes across the United States and to accurately predict future incidents.

The tool, called SafeTraffic Copilot, aims to provide experts with both crash analyses and crash predictions to reduce the rising number of fatalities and injuries that happen on U.S. roads each year.

The work, led by Johns Hopkins University researchers, is published in Nature Communications.

New Technique Auto-Selects Training Examples to Speed Up Fine-Tuning

Fine-tuning large language models via reinforcement learning is computationally expensive, but researchers found a way to streamline the process.

What’s new: Qinsi Wang and colleagues at UC Berkeley and Duke University developed GAIN-RL, a method that accelerates reinforcement learning fine-tuning by selecting training examples automatically based on the model’s own internal signals, specifically the angles between vector representations of tokens. The code is available on GitHub.

Key insight: The cosine similarity between a model’s vector representations of input tokens governs the magnitude of gradient updates during training. Specifically, the sum of those similarities that enter a model’s classification layer, called the angle concentration, governs the magnitude of gradient updates. Examples with higher angle concentration produce larger gradient updates. The magnitude of a gradient update in turn determines the effectiveness of a given training example: The larger the update, the more the model learns. Prioritizing the most-effective examples before transitioning to less-effective ones enhances training efficiency while adding little preprocessing overhead.

Novel method for controlling Faraday rotation in conductive polymers

Researchers at the University of Tsukuba have developed a novel method for controlling the optical rotation of conductive polymer polythiophene in a magnetic field at low voltage. This method combines the “Faraday rotation” phenomenon, in which a polarizing plane rotates in response to a magnetic field, with the electrochemical oxidation and reduction of conductive polymers.

The study is published in the journal Molecular Crystals and Liquid Crystals.

Conductive polymers possess various properties in addition to conductivity, with applications in light-emitting devices, electromagnetic wave shielding, and anticorrosion materials.

View a PDF of the paper titled Vision-Zero: Scalable VLM Self-Improvement via Strategic Gamified Self-Play, by Qinsi Wang and 8 other authors

Although reinforcement learning (RL) can effectively enhance the reasoning capabilities of vision-language models (VLMs), current methods remain heavily dependent on labor-intensive datasets that require extensive manual construction and verification, leading to extremely high training costs and consequently constraining the practical deployment of VLMs. To address this challenge, we propose Vision-Zero, a domain-agnostic framework enabling VLM self-improvement through competitive visual games generated from arbitrary image pairs. Specifically, Vision-Zero encompasses three main attributes: Strategic Self-Play Framework: Vision-Zero trains VLMs in “Who Is the Spy”-style games, where the models engage in strategic reasoning and actions across multiple roles. Through interactive gameplay, models autonomously generate their training data without human annotation. Gameplay from Arbitrary Images: Unlike existing gamified frameworks, Vision-Zero can generate games from arbitrary images, thereby enhancing the model’s reasoning ability across diverse domains and showing strong generalization to different tasks. We demonstrate this versatility using three distinct types of image datasets: CLEVR-based synthetic scenes, charts, and real-world images. Sustainable Performance Gain: We introduce Iterative Self-Play Policy Optimization (Iterative-SPO), a novel training algorithm that alternates between Self-Play and reinforcement learning with verifiable rewards (RLVR), mitigating the performance plateau often seen in self-play-only training and achieving sustained long-term improvements. Despite using label-free data, Vision-Zero achieves state-of-the-art performance on reasoning, chart question answering, and vision-centric understanding tasks, surpassing other annotation-based methods. Models and code has been released at https://github.com/wangqinsi1/Vision-Zero.

New Sodium Battery Design Works Even at Subzero Temperatures

A new technique stabilizes a metastable form of sodium solid electrolyte, enabling all-solid-state sodium batteries to maintain performance even at subzero temperatures. All-solid-state batteries are considered a safe and powerful option for running electric vehicles, electronics, and even storin

How diamond fails under extreme electrical fields

A research team from the University of Chinese Academy of Sciences has revealed the failure mechanism of diamond under extreme electrical fields through in situ experiments and molecular dynamics simulations. The study, published in Cell Reports Physical Science, provides critical insights for the design of robust diamond devices.

Diamond is known for its exceptional physical properties, including ultra-high breakdown field strength and , making it a promising material for and high-power electronics. However, its failure process under extreme electrical fields has remained poorly understood before now.

Led by Profs. Yan Qingbo and Chen Guangchao, the researchers used an in situ transmission electron microscopy (TEM) method to observe the breakdown process in real time. They found that diamond failure begins preferentially along the (111) crystal plane due to stress-induced lattice distortion and subsequent amorphization, rather than transforming into graphite.

Microsoft’s new AI feature will organize your photos automatically

Microsoft has begun testing a new AI-powered feature in Microsoft Photos, designed to categorize photos automatically on Windows 11 systems.

Dubbed Auto-Categorization, it is currently limited to sorting screenshots, receipts, identity documents, and notes, and it’s rolling out to Copilot+ PCs across all Windows Insider channels with Microsoft Photos version 2025.11090.25001.0 or higher.

Microsoft says the feature utilizes a language-agnostic AI model that identifies document types regardless of the language used in the image. It works by grouping photos into predefined folders automatically, based on their visual content, such as handwritten notes, receipts, or printed documents.

CWM: An Open-Weights LLM for Research on Code

We release Code World Model (CWM), a 32-billion-parameter open-weights LLM, to advance research on code generation with world models. To improve code understanding beyond what can be learned from training on static code alone, we mid-train CWM on a large amount of observation-action trajectories from Python interpreter and agentic Docker environments, and perform extensive multi-task reasoning RL in verifiable coding, math, and multi-turn software engineering environments. With CWM, we provide a strong testbed for researchers to explore the opportunities world modeling affords for improving code generation with reasoning and planning in computational environments. We present first steps of how world models can benefit agentic coding, enable step-by-step simulation of Python code execution, and show early results of how reasoning can benefit from the latter.

/* */