Researchers found that large language models use a semantic hub mechanism to process diverse data, integrating information from various modalities. They demonstrated that models can change their outputs by intervening with text in the dominant language, even when processing other languages.
MIT engineers have developed a fully 3D-printed electrospray engine that can be produced rapidly and at a lower cost than traditional thrusters. The device generates stable and uniform flow of propellant, producing as much or more thrust than existing droplet-emitting electrospray engines.
MIT researchers have introduced a new system called MiFly that enables drones to self-localize in indoor, dark, and low-visibility environments. The system uses radio frequency waves reflected by a single tag placed in the environment, allowing the drone to estimate its trajectory with high accuracy.
Astronomers have discovered a key ingredient within the Phoenix cluster that explains its mysterious starburst, using NASA's James Webb Space Telescope to observe the cluster. The detection of warm gas confirms that the Phoenix cluster is actively cooling and able to generate a huge amount of stellar fuel on its own.
MIT researchers have developed a new approach to assess predictions with a spatial dimension, leading to more accurate forecasts in spatial prediction tasks. The new method provides better validations than classical methods, especially in cases where assumptions about data independence break down.
A new study shows Indian students who work in retail markets perform poorly on school-type math problems, while those still enrolled in school do better on marketplace math. This contrast highlights a need to address the skills gap between academic and applied math, with experts suggesting different approaches for each group
Researchers at MIT and Harvard University have directly measured superfluid stiffness in magic-angle graphene for the first time, shedding light on its remarkable properties. The study suggests that quantum geometry governs the material's superconductivity, a key step toward understanding its exceptional properties.
A team of MIT engineers has developed a training method for multiagent systems that can guarantee their safe operation in crowded environments. The method enables agents to continually map their safety margins, allowing them to scale up to any number of agents while maintaining system safety.
Researchers created an automated system that enables developers to build algorithms taking advantage of two types of data redundancy, boosting computation speed and reducing energy consumption. The system can optimize machine-learning algorithms for a wide range of applications, including scientific computing.
Researchers use generative AI to predict chromatin structures in single cells, overcoming limitations of existing experimental methods. The technique can generate thousands of structure predictions in minutes, enabling faster study of how 3D genome organization affects gene expression.
Researchers found that training AI agents in a less noisy environment can lead to better performance than traditional methods. The indoor training effect suggests that constructing simulated environments with specific noise levels can improve AI learning.
Researchers developed a nanoparticle-based vaccine that generates antibodies targeting conserved regions of sarbecovirus receptor-binding proteins, offering broader protection against multiple strains. The vaccine demonstrated strong antibody responses and protection in animal studies against diverse SARS-CoV-2 and other sarbecoviruses.
A new computational model reveals how place cells in the hippocampus can store episodic memories, including those of events without a spatial component. The model proposes that grid cells form a scaffold that anchors memories and drives recall by connecting to sensory cortex.
Researchers at MIT have developed a new design for robotic insects that can perform precise pollination with increased speed and maneuverability. The revamped robot has a longer flight duration and can complete acrobatic maneuvers, enabling it to aid in mechanical pollination and boost fruit yields.
Astronomers have observed the same supermassive black hole exhibiting unprecedented behavior, with X-ray flashes increasing in frequency over two years. The source of these flashes is believed to be a spinning white dwarf core orbiting around the black hole's edge.
A study by MIT researchers shows that households in Amsterdam reduced their energy consumption and costs by half after receiving better information and coaching. The experiment found that smart devices can help people change their habits, but only if accompanied by a close understanding of what drives behavior changes.
MIT engineers developed a nanofiltration process to capture aluminum ions from cryolite waste, reducing hazardous waste and improving efficiency. The membrane selectively captured over 99% of aluminum ions, enabling the recovery of aluminum and reducing the need for new mining.
Physicists at MIT have devised a formula predicting which invasive species will succeed in an ecosystem, based on analysis of soil bacteria. The researchers found that population fluctuations are key to invasion success, and that diversity can either help or hinder the invader.
Astronomers at MIT used a novel technique to determine the precise location of a fast radio burst, finding it likely originated from the magnetosphere around a rotating neutron star. The study provides conclusive evidence that fast radio bursts can emerge from this highly magnetic environment.
MIT researchers developed a biosensing technique that eliminates the need for wires, using tiny wireless antennas with light detection to measure electrical signals from cells. The devices can capture scattered light with an optical microscope and measure signals with micrometer spatial resolution.
Researchers create SciAgents framework to autonomously generate and evaluate promising research hypotheses in biologically inspired materials. The framework uses graph reasoning methods to organize relationships between scientific concepts, mimicking biological systems.
Researchers create multilayered chip design that doesn't require silicon wafer substrates, allowing for better communication and computation between layers. This breakthrough enables the construction of fast and powerful AI hardware comparable to supercomputers.
Researchers at MIT have created a new magnetic state in an antiferromagnetic material using terahertz laser light, enabling controlled switching and potentially leading to more efficient memory chips. The technique provides a powerful tool for manipulating magnetism and advancing information processing technology.
A study by MIT economists suggests that industrial policy has a relatively limited impact on GDP growth, with average long-run gains of 1.08% in favorable circumstances. The research indicates that local consumer demand can curb the impact of industrial policy, putting a ceiling on directed growth.
Researchers at MIT have developed Boltz-1, an open-source AI model that achieves state-of-the-art performance in predicting biomolecular structures. The model surpasses AlphaFold3, which is limited to academic research and commercial use, by incorporating new algorithms and improving prediction efficiency.
A new noninvasive imaging method developed by MIT researchers can penetrate deeper into living tissue than previous techniques, producing richer and more detailed images. This breakthrough enhances biological research capabilities, enabling scientists to study immune responses and develop new medicines with greater accuracy.
A new technique identifies and removes specific points in a training dataset that contribute most to a model's failures on minority subgroups. This approach maintains the overall accuracy of the model while improving its performance regarding underrepresented groups.
A new system, EXPLINGO, enables AI models to generate readable narratives explaining their predictions, helping users make better decisions. The system, developed by MIT researchers, uses large language models to transform complex explanations into plain language and automatically evaluate their quality.
A new detection method has been used to identify over 100 small asteroids in the main asteroid belt, ranging from bus-sized to stadium-wide. This breakthrough could aid in tracking potential asteroid impactors and provide insights into meteorite origins.
Researchers at MIT have developed a class of biodegradable materials that can replace plastic beads used in some health and beauty products. The particles are shown to be effective in encapsulating essential nutrients such as vitamin A, which could help alleviate nutrient deficiencies in populations worldwide.
A new study shows that coordinating the siting of solar farms, wind farms, and storage systems can improve efficiency and reduce costs by aligning renewable power generation with demand. The approach uses fine-grained weather data and energy system modeling to maximize utilization of renewable resources.
Researchers develop a simple fix to an existing technique, enabling the generation of sharp, high-quality 3D shapes that rival top model-generated 2D images. The new approach improves upon previous methods by avoiding costly retraining and complex postprocessing.
A new catalyst converts methane into polymers at room temperature and atmospheric pressure, making it easier to deploy at sites of methane production. The catalyst also enables the creation of sealants to heal cracks in natural gas pipes, potentially reducing methane leakage.
A new type of gastric balloon has been designed to help people lose weight by making them feel too full to overeat. The balloon can be inflated and deflated as needed, and animal studies have shown that it can reduce food intake by 60 percent before meals.
Scientists at MIT developed a fully integrated photonic processor that can perform all key computations of a deep neural network optically on the chip. The device completed machine-learning classification tasks in under half a nanosecond while achieving over 92% accuracy, similar to traditional hardware.
A new AI tool generates realistic satellite images of future flooding, which can help communities visualize and prepare for approaching storms. The method combines a generative artificial intelligence model with a physics-based flood model, producing more accurate and realistic images than an AI-only approach.
The MIT team fabricated a simple water filter modeled after the mobula ray's plankton-filtering features and studied its performance. They found that the ray's filtering features are broadly similar to industrial cross-flow filters, which could inform design of water treatment systems.
Researchers at MIT have introduced a new algorithm that strategically selects the best tasks for training an AI agent, resulting in improved performance and reduced training costs. The technique outperforms existing methods by five to 50 times, making it more efficient and effective.
A study published in Psychological Science found that 2-year-olds rely on grammatical information to understand the meaning of new words. The researchers used focus signals in sentences to manipulate children's responses, showing that young kids can absorb grammatical cues and leverage them to learn new words.
Lightman's book delves into the emotional and intellectual connections we make with natural wonders, using examples like aurora borealis and fall foliage. He grounds his grasp of nature in scientific explanation while still finding awe and wonder.
Researchers developed an ingestible capsule that releases a burst of drugs into the stomach or other organs, eliminating the need for injections. The capsule uses compressed carbon dioxide or tightly coiled springs to propel liquid drugs out, targeting specific parts of the digestive tract.
Physicists at MIT have made a breakthrough discovery that sheds light on the conditions that lead to exotic electronic states in graphene and other two-dimensional systems. Through calculations, they show that pentalayer graphene can exhibit fractional charge without a magnetic field.
Researchers created a new electrode design that increases the efficiency of converting CO2 into ethylene, a valuable chemical product. The electrochemical system can now be scaled up for industrial applications without significant energy or cost losses.
A new study by MIT engineers reveals that exercise can stimulate nerve growth, with neurons growing four times farther in the presence of myokines released during muscle contractions. Physical effects of exercise, such as repeated stretching and pulling, also promote nerve growth, challenging previous biochemical-only theories.
MIT researchers developed a theoretical foundation for methods to aggregate genes into related groups, enabling the efficient learning of underlying cause-and-effect relationships. They achieve this using only observational data, potentially paving the way for targeted treatments.
Scientists analyzed particles from asteroid Ryugu, revealing a weak magnetic field that likely pulled matter inward to form the outer planetary bodies. The team estimates that such a low-grade field intensity would have been enough to play a role in giant planet formation, from Jupiter to Neptune.
Despite their ability to provide turn-by-turn driving directions in New York City with near-perfect accuracy, generative AI models do not form a coherent internal map of the city. The researchers developed new metrics to test a transformer's world model and found that only one generated a coherent world model for Othello moves.
A team of MIT engineers has developed a new computational method for analyzing complex biological systems, including the immune system's response to tuberculosis vaccination. The approach uses probabilistic graphical networks to identify key interactions and mechanisms, shedding light on how vaccines induce immunity.
MIT researchers develop 3D transistors using quantum mechanical properties to achieve low-voltage operation and high performance. The devices can deliver comparable performance to state-of-the-art silicon transistors while operating efficiently at much lower voltages.
A team of MIT engineers developed an algorithm to identify causal links in complex systems, taking data from various sources and analyzing interactions between variables. The method generates a causality map linking variables with likely cause-and-effect relationships, including synergistic and redundant links.