Toggle light / dark theme

Researchers confront a formidable challenge within the expansive domain of materials science—efficiently distilling essential insights from densely packed scientific texts. This intricate dance involves navigating complex content and generating coherent question-answer pairs that encapsulate the core of the material. The complexity lies in the substantial task of extracting pivotal information from the dense fabric of scientific texts, requiring researchers to craft meaningful question-answer pairs that capture the essence of the material.

Current methodologies within this domain often lean on general-purpose language models for information extraction. However, these approaches need help with text refinement and the accurate incorporation of equations. In response, a team of MIT researchers introduced MechGPT, a novel model grounded in a pretrained language model. This innovative approach employs a two-step process, utilizing a general-purpose language model to formulate insightful question-answer pairs. Beyond mere extraction, MechGPT enhances the clarity of key facts.

The journey of MechGPT commences with a meticulous training process implemented in PyTorch within the Hugging Face ecosystem. Based on the Llama 2 transformer architecture, the model flaunts 40 transformer layers and leverages rotary positional embedding to facilitate extended context lengths. Employing a paged 32-bit AdamW optimizer, the training process attains a commendable loss of approximately 0.05. The researchers introduce Low-Rank Adaptation (LoRA) during fine-tuning to augment the model’s capabilities. This involves integrating additional trainable layers while freezing the original pretrained model, preventing the model from erasing its initial knowledge base. The result is heightened memory efficiency and accelerated training throughput.

An advancement in neutron shielding, a critical aspect of radiation protection, has been achieved. This breakthrough is poised to revolutionize the neutron shielding industry by offering a cost-effective solution applicable to a wide range of materials surfaces.

A research team, led by Professor Soon-Yong Kwon in the Graduate School of Semiconductors Materials and Devices Engineering and the Department of Materials Science and Engineering at UNIST has successfully developed a neutron shielding film capable of blocking neutrons present in radiation. This innovative shield is not only available in large areas but also lightweight and flexible.

The team’s paper is published in the journal Nature Communications.

The simple story line that ‘Gell-Mann and Zweig invented quarks in 1964 and the quark model was generally accepted after 1968 when deep inelastic electron scattering experiments at SLAC showed that they are real’ contains elements of the truth, but is not true. This paper describes the origins and development of the quark model until it became generally accepted in the mid-1970s, as witnessed by a spectator and some-time participant who joined the field as a graduate student in October 1964. It aims to ensure that the role of Petermann is not overlooked, and Zweig and Bjorken get the recognition they deserve, and to clarify the role of Serber.

This is almost like endowing a printer with a set of eyes and a brain, where the eyes observe what is being printed, and then the brain of the machine directs it as to what should be printed next.


Moritz Hocher.

Traditional systems use nozzles to deposit tiny drops of resin, smoothed over with a scraper or roller and then curved with UV light. However, this smoothing limits the materials that could be used since slow-curing resins could be squished or smeared.

New materials developed at the University of Surrey could pave the way for a new generation of flexible X-ray detectors, with potential applications ranging from cancer treatment to better airport scanners.

Traditionally, X-ray detectors are made of heavy, rigid material such as silicon or germanium. New, flexible detectors are cheaper and can be shaped around the objects that need to be scanned, improving accuracy when screening patients and reducing risk when imaging tumors and administering radiotherapy.

Dr. Prabodhi Nanayakkara, who led the research at the University of Surrey, said, “This new material is flexible, low-cost, and sensitive. But what’s exciting is that this material is tissue equivalent. This paves the way for live dosimetry, which just isn’t possible with current technology.”

“This work brings us a step closer to realizing the full potential of physical reservoirs to create computers that not only require significantly less energy, but also adapt their computational properties to perform optimally across various tasks, just like our brains,” said Dr. Oscar Lee.


A recent study published in Nature Materials examines a breakthrough approach in physical reservoir computing, also known as a neuromorphic or brain-inspired method and involves using a material’s physical properties to adhere to a myriad of machine learning duties. This study was conducted by an international team of researchers and holds the potential to help physical reservoir computing serve as a framework towards making machine learning more energy efficient.

Artist rendition of connected chiral (twisted) magnets used as a computing avenue for brain-inspired, physical reservoir computing. (Credit: Dr. Oscar Lee)

For the study, the researchers used a magnetic field and temperature variances on chiral (twisted) magnets—which served as the computing channel—they found the materials could be used for a myriad of machine learning needs. What makes this discovery extraordinary is that physical reservoir computing has been found to have limits, specifically pertaining to its ability to be rearranged. Additionally, the team discovered that the chiral magnets performed better at certain computing tasks based on changes in the magnetic field phases used throughout the experiments.