Aurelia Butler

Large language models (LLMs) — the advanced AI behind tools like ChatGPT — are increasingly integrated into daily life, assisting with tasks such as writing emails, answering questions, and even supporting healthcare decisions. But can these models collaborate with others in the same way humans do? Can they understand social situations, make compromises, or establish trust? A new study from researchers at Helmholtz Munich, the Max Planck Institute for Biological Cybernetics, and the University of Tübingen, reveals that while today’s AI is smart, it still has much to learn about social intelligence.
Playing Games to Understand AI Behavior
To find out how LLMs behave in social situations, researchers applied behavioral game theory — a method typically used to study how people cooperate, compete, and make decisions. The team had various AI models, including GPT-4, engage in a series of games designed to simulate social interactions and assess key factors such as fairness, trust, and cooperation.
The researchers discovered that GPT-4 excelled in games demanding logical reasoning — particularly when prioritizing its own interests. However, it struggled with tasks that required teamwork and coordination, often falling short in those areas.
“In some cases, the AI seemed almost too rational for its own good,” said Dr. Eric Schulz, lead author of the study. “It could spot a threat or a selfish move instantly and respond with retaliation, but it struggled to see the bigger picture of trust, cooperation, and compromise.”
Teaching AI to Think Socially
To encourage more socially aware behavior, the researchers implemented a straightforward approach: they prompted the AI to consider the other player’s perspective before making its own decision. This technique, called Social Chain-of-Thought (SCoT), resulted in significant improvements. With SCoT, the AI became more cooperative, more adaptable, and more effective at achieving mutually beneficial outcomes — even when interacting with real human players.

“Once we nudged the model to reason socially, it started acting in ways that felt much more human,” said Elif Akata, first author of the study. “And interestingly, human participants often couldn’t tell they were playing with an AI.”
Applications in Health and Patient Care
The implications of this study reach well beyond game theory. The findings lay the groundwork for developing more human-centered AI systems, particularly in healthcare settings where social cognition is essential. In areas like mental health, chronic disease management, and elderly care, effective support depends not only on accuracy and information delivery but also on the AI’s ability to build trust, interpret social cues, and foster cooperation. By modeling and refining these social dynamics, the study paves the way for more socially intelligent AI, with significant implications for health research and human-AI interaction.
“An AI that can encourage a patient to stay on their medication, support someone through anxiety, or guide a conversation about difficult choices,” said Elif Akata. “That’s where this kind of research is headed.” More

Numbers have a funny way about them. Young math students are taught various strategies to make problem-solving easier. Comparing fractions? Find a common denominator or convert to decimals. The strategies get more complex when doing the kind of math used to describe the activities of DNA, RNA, or protein sequences.
In science, when you make a model, its parameters determine its predictions. But what do you do when different sets of parameters result in the same predictions? Call one half 2/4 or 3/6 — either way, the result’s the same. In physics, such parameter sets are called gauge freedoms. They play a key role in how we understand electromagnetism and quantum mechanics. Surprisingly, gauge freedoms also arise in computational biology when trying to model how different mutations interact.
Now, Cold Spring Harbor Laboratory (CSHL) quantitative biologists have developed a unified theory for gauge freedoms in models of biological sequences. Their solution could have countless applications, from plant breeding to drug development.
Granted, most folks have never heard of gauge freedoms. So, how common are they? When it comes to computer models used to describe massive genetic datasets, they’re basically everywhere, says CSHL Associate Professor Justin Kinney, who co-led this study with Associate Professor David McCandlish.
“Gauge freedoms are ubiquitous in computational models of how biological sequences work,” Kinney says. “Historically, they’ve been dealt with as annoying technicalities. We’re the first to study them directly in order to get a deeper understanding of where they come from and how to handle them.”
Until now, computational biologists have accounted for gauge freedoms using a variety of ad hoc approaches. Kinney, McCandlish, and their colleagues were looking for a better way. Together, they developed a unified approach. Their new mathematical theory provides efficient formulas scientists can use for all sorts of biological applications. These formulas will allow scientists to interpret research results much faster and with greater confidence.
The investigators also published a companion paper that reveals where these gauge freedoms ultimately come from. It turns out they’re needed for models to reflect symmetries in real biological sequences. Perhaps counterintuitively, making biological models behave in a simple and intuitive way requires them to be larger and more complex. “We prove that gauge freedoms are necessary to interpret the contributions of particular genetic sequences,” McCandlish adds.
Together, the studies strongly suggest that Kinney and McCandlish’s unified approach isn’t just a new strategy for solving theoretical problems. It may prove fundamental for future efforts in agriculture, drug discovery, and beyond. More

A team of physicists at the University of Cambridge has unveiled a breakthrough in quantum sensing by demonstrating the use of spin defects in Hexagonal Boron Nitride (hBN) as powerful, room-temperature sensors capable of detecting vectorial magnetic field at the nanoscale. The findings, published in Nature Communications, mark a significant step toward more practical and versatile quantum technologies.
“Quantum sensors allow us to detect nanoscale variations of various quantities. In the case of magnetometry, quantum sensors enable nanoscale visualisation of properties like current flow and magnetisation in materials leading to the discovery of new physics and functionality,” said Dr Carmem Gilardoni, co-first author of this study at Cambrdge’s Cavendish Laboratory. “This work takes that capability to the next level using hBN, a material that’s not only compatible with nanoscale applications but also offers new degrees of freedom compared to state-of-the-art nanoscale quantum sensors.”
To date, nanoscale quantum magnetometry at ambient conditions is only possible with the nitrogen vacancy (NV) centre defect in diamond. While a powerful technology, these sensors have limitations that result from their fundamental photophysics. In particular, the NV centre is a single-axis sensor, with limited dynamic range for magnetic field detection. In contrast, the hBN sensor development by the team in Cambridge does not share these limitations and instead presents a multi-axis sensor of magnetic field with large dynamic range.
The team’s work demonstrates the capabilities of this new sensor, as well as providing a mechanistic understanding of the origin of its advantageous properties for sensing. Importantly, the team uncovered that the low symmetry, and fortuitous excited state optical rates are responsible for the dynamic range and vectorial capabilities.
hBN is a two-dimensional material, similar to graphene, that can be exfoliated to just a few atomic layers thick. Atomic-scale defects in the hBN lattice absorb and emit visible light in a way that is sensitive to local magnetic conditions, making it an ideal candidate for quantum sensing applications.
In this study, the team investigated the response of the hBN defect fluorescence to variations in magnetic field, using a technique known as optically detected magnetic resonance (ODMR). By carefully tracking the spin response and combining this with detailed analysis of the dynamics of photon emission, the team could uncover the underlying optical rates of the system and their connection to the defect symmetry, and how this combination results in a robust and versatile magnetic field sensor.
“ODMR isn’t a new technique — but what we have shown is that probes built using the hBN platform would allow this technique to be applied in a variety of new situations. It’s exciting because it opens the door to imaging magnetic phenomena and nanomaterials in a way we couldn’t before,” said Dr Simone Eizagirre Barker, co-first author of the paper.
“This sensor could open the door to studying magnetic phenomena in new material systems, or with higher spatial resolution that done before,” said Prof Hannah Stern, who co-led the research with Prof Mete Atatüre at the Cavendish Laboratory. “The 2D nature of the host material also opens exciting new possibilities for using this sensor. For example, the spatial resolution for this technique is determined by the distance between the sample and sensor. With an atomically-thin material, we can potentially realise atomic scale spatial mapping of magnetic field.” More

Recent research points to the potential utility of a familiar sounding foe-herpes virus-in the fight against cancer.
The idea: the virus has evolved to commandeer cellular machinery in order activate signaling pathways inside cells and these strategies can be repurposed to bolster immunotherapy against diseases like cancer.
T cells are front line defenders against pathogens, like viruses, and cancer because they can kill infected or malignant cells.
Scientists have for years been trying different techniques to direct these immune cells to protect against disease.
CAR-T therapy is one such example of prompting the body’s own immune system to attack certain forms of cancer using T cells.
However, the therapeutic potential of T cells can be limited by the suppressive environment present within tumors that impairs T cell survival and function.
The University of Michigan team identified herpes virus saimiri, which infects the T cells of squirrel monkeys, as a source of proteins that activate pathways in T cells that are needed to promote T cell survival.

The work, led by the lab of Adam Courtney, Ph.D., in the Department of Pharmacology and the U-M Rogel Cancer Center, exploits this ability in order to investigate whether a modified viral protein could be used to activate transcription factors known as STAT proteins.
The approach is borne of observations that stimulation of the JAK-STAT5 pathway by cytokines like interleukin-2 (IL-2) helps boost the therapeutic ability of T cells to kill cancer cells.
The team engineered a variant of the tyrosine kinase interacting protein from the herpes virus to bind LCK (a kinase active in resting T cells) and recruit it to activate STAT5.
In this way, the team determined that direct activation of STAT5 could sustain T cell function in tumors of mouse models of melanoma and lymphoma.
Their findings hint at a new approach — using genes from organisms with proven ability to modulate human cells — to enhance the power of immunotherapy.
Ph.D. candidate Yating Zheng, of the Department of Pharmacology at U-M Medical School is first author of the paper. More

In a major leap forward for genetic and biomedical research, two scientists at the University of Missouri have developed a powerful new artificial intelligence tool that can predict the 3D shape of chromosomes inside individual cells — helping researchers gain a new view of how our genes work.
Chromosomes are the tiny storage boxes that hold our DNA. Since each cell has about six feet of DNA packed inside it, it must be folded up tightly to fit. This folding not only saves space — it also controls which genes are active or inactive. But when the DNA doesn’t fold the right way, it can disrupt normal cell functions and lead to serious diseases, including cancer.
Historically, scientists have relied on data that averaged results from millions of cells at once. That makes it almost impossible to see the unique differences between individual cells. But the new AI model developed by Yanli Wang and Jianlin “Jack” Cheng at Mizzou’s College of Engineering changes that.
“This is important because even cells from the same part of the body can have chromosomes folded in very different ways,” Wang, a graduate student and lead author of the study, said. “That folding controls which genes are turned on or off.”
Studying single cells is tricky because the data is often messy or incomplete. But the new AI tool is specially designed to work with those challenges. It’s smart enough to spot weak patterns in noisy data, and it knows how to estimate a chromosome’s 3D shape even when some information is missing.
It also understands how to “see” biological structures correctly, even when they’re rotated. Compared to a previous deep learning AI method, Mizzou’s tool is more than twice as accurate when analyzing human single-cell data.
The team has made the software free and available to scientists around the world. That means researchers can now use it to better understand how genes function, how diseases start and how to design better treatments.
“Every single cell can have a different chromosome structure,” Cheng, a Curators’ Distinguished Professor of Electrical Engineering and Computer Science, said. “Our tool helps scientists study those differences in detail — which can lead to new insights into health and disease.”
The researchers now plan to improve the AI tool even further by expanding it to build the high-resolution structures of entire genomes. Their goal: to give scientists the clearest picture yet of the genetic blueprint inside our cells. More

A new study in Nature describes both the mechanism and the material conditions necessary for superfluorescence at room temperature. The work could serve as a blueprint for designing materials that allow exotic quantum states — such as superconductivity, superfluidity or superfluorescence — at high temperatures, paving the way for applications such as quantum computers that don’t require extremely low temperatures to operate.
The international team that did the work was led by North Carolina State University and included researchers from Duke University, Boston University and the Institut Polytechnique de Paris.
“In this work, we show both experimental and theoretical reasons behind macroscopic quantum coherence at high temperature,” says Kenan Gundogdu, professor of physics at NC State and corresponding author of the study. “In other words, we can finally explain how and why some materials will work better than others in applications that require exotic quantum states at ambient temperatures.”
Picture a school of fish swimming in unison or the synchronized flashing of fireflies — examples of collective behavior in nature. When similar collective behavior happens in the quantum world — a phenomenon known as macroscopic quantum phase transition — it leads to exotic processes such as superconductivity, superfluidity, or superfluorescence. In all these processes a group of quantum particles forms a macroscopically coherent system that acts like a giant quantum particle.
However, quantum phase transitions normally require super cold, or cryogenic, conditions to occur. This is because higher temperatures create thermal “noise” that disrupts the synchronization and prevents the phase transition.
In a previous study, Gundogdu and colleagues had determined that the atomic structure of some hybrid perovskites protected the groups of quantum particles from the thermal noise long enough for the phase transition to occur. In these materials, large polarons — groups of atoms bound to electrons — formed, insulating light emitting dipoles from thermal interference and allowing superfluorescence.
In the new study, the researchers found out how the insulating effect works. When they used a laser to excite the electrons within the hybrid perovskite they studied, they saw large groups of polarons coming together. This grouping is called a soliton.

“Picture the atomic lattice as a fine cloth stretched between two points,” Gundogdu says. “If you place solid balls — which represent excitons — on the cloth, each ball deforms the cloth locally. To get an exotic state like superfluorescence you need all the excitons, or balls, to form a coherent group and interact with the lattice as a unit, but at high temperatures thermal noise prevents this.
“The ball and its local deformation together form a polaron,” Gundogdu continues. “When these polarons transition from a random distribution to an ordered formation in the lattice, they make a soliton, or coherent unit. The soliton formation process dampens the thermal disturbances, which otherwise impede quantum effects.”
“A soliton only forms when there is enough density of polarons excited in the material,” says Mustafa Türe, NC State Ph.D. student and co-first author of the paper. “Our theory shows that if the density of polarons is low, the system has only free incoherent polarons, whereas beyond a threshold density, polarons evolve into solitons.”
“In our experiments we directly measured the evolution of a group of polarons from an incoherent uncorrelated phase to an ordered phase,” adds Melike Biliroglu, postdoctoral researcher at NC State and co-first author of the work. “This is one of the first direct observations of macroscopic quantum state formation.”
To confirm that the soliton formation suppresses the detrimental effects of temperature, the group worked with Volker Blum, the Rooney Family Associate Professor of Mechanical Engineering and Materials Science at Duke, to calculate the lattice oscillations responsible for thermal interference. They also collaborated with Vasily Temnov, professor of physics at CNRS and Ecole Polytechnique, to simulate the recombination dynamics of the soliton in the presence of thermal noise. Their work confirmed the experimental results and verified the intrinsic coherence of the soliton.
The work represents a leap forward in understanding both how and why certain hybrid perovskites are able to exhibit exotic quantum states.
“Prior to this work it wasn’t clear if there was a mechanism behind high temperature quantum effects in these materials,” says Franky So, co-author of the paper and the Walter and Ida Freeman Distinguished Professor of Materials Science and Engineering at NC State.
“This work shows a quantitative theory and backs it up with experimental results,” Gundogdu says. “Macroscopic quantum effects such as superconductivity are key to all the quantum technologies we are pursuing — quantum communication, cryptology, sensing and computation — and all of them are currently limited by the need for low temperatures. But now that we understand the theory, we have guidelines for designing new quantum materials that can function at high temperatures, which is a huge step forward.”
The work is supported by the Department of Energy, Office of Science (grant no. DE-SC0024396). Researchers Xixi Qin, and Uthpala Herath from Duke University; Anna Swan from Boston University; and Antonia Ghita from the Institut Polytechnique de Paris, also contributed to the work. More

As fast as modern electronics have become, they could be much faster if their operations were based on light, rather than electricity. Fiber optic cables already transport information at the speed of light; to do computations on that information without translating it back to electric signals will require a host of new optical components.
Engineering researchers at the University of Utah have now developed such a device — one that can be adjusted on the fly to give light different degrees of circular polarization. Because information can be stored in a property of light known as chirality, the researchers’ device could serve as a multifunctional, reconfigurable component of an optical computing system.
Led by Weilu Gao, assistant professor in the Department of Electrical & Computer Engineering, and Jichao Fan, a Ph.D. candidate in his lab at the John and Marcia Price College of Engineering, a study demonstrating the device was published in the journal Nature Communications.
Chiral light refers to electromagnetic waves that exhibit handedness; they can be either left-handed or right-handed. This “handedness” arises from the rotation of the magnetic fields as the light propagates, creating a spiral structure.
“Traditional chiral optics were like carved stone — beautiful but frozen,” Gao said. “This made them not useful for applications requiring real-time control, like reconfigurable optical computing or adaptive sensors.”
“We’ve created ‘living’ optical matter that evolves with electrical pulses,” Fan said, “thanks to our aligned-carbon-nanotube-phase-change-material heterostructure that merges light manipulation and memory into a single scalable platform.”
This “heterostructure” consists of a stack of multiple different thin films, including a collection of aligned carbon nanotubes with different orientations. Other films in the stack consist of germanium-antimony-tellurium, a well-known “phase-change material” or PCM. An electrical pulse along the carbon nanotube layer introduces heat, which in turn causes the PCM layer’s internal structure to transition from amorphous to crystalline.

“The carbon nanotubes simultaneously act as chiral optical elements and transparent electrodes for PCM switching — eliminating the need for separate control components,” Fan said.
Critically, this change modifies the heterostructure’s circular dichroism, which means it can be made to absorb different types of circularly polarized light at different strengths. The research team’s advances in manufacturing techniques and artificial-intelligence-assisted design enabled these layers to be assembled into a stacked heterostructure without degrading their individual optical properties.
Once assembled, the layers selectively reduce the amount of left- or right-circularly polarized light that passes through them, depending on the state of the PCM layer. And because that phase change can be initiated by an electrical pulse, the structure’s overall circular dichroism can be adjusted in real-time.
The researchers were able to achieve this on the wafer-scale, because of the scalable manufacturing of aligned carbon nanotubes and phase-change-material films.
Being able to modify the device’s circular dichroism gives researchers fine-grained control over which direction circularly polarized light twists, meaning its “handedness” can be used as memory in an optical circuit. In addition to light’s speed advantage over electricity, there are additional properties of light in which information can be stored in parallel.
“By adding circular dichroism as an independent parameter, we create an orthogonal information channel,” Gao said. “Adjusting it does not interfere with other properties like amplitude or wavelength.”
The research was supported by the National Science Foundation through Grants No. 2230727, No. 2235276, No. 2316627 and No. 2321366. More

Learning to play music by ear is challenging for most musicians, but research from a team at the University of Waterloo may help musicians-in-training find the right notes.
The Waterloo team analyzed a range of YouTube videos that focused on learning music by ear and identified four simple ways music learning technology can better aid prospective musicians — helping people improve recall while listening, limiting playback to small chunks, identifying musical subsequences to memorize, and replaying notes indefinitely.
“There are a lot of apps and electronic tools out there to help learn by ear from recorded music,” said Christopher Liscio, a recent Waterloo master’s graduate in computer science and the study’s lead author.
“But we see evidence that musicians don’t appear to use them very much, which makes us question whether these tools are truly well-suited to the task. By studying how people teach and learn how to play music by ear in YouTube videos, we can try to understand what might actually help these ear-learning musicians.”
The team studied 28 YouTube ear-learning lessons, breaking each down to examine how the instructors structured their teaching and how students would likely retain what they heard. Surprisingly, they found that very few creators or viewers were using existing digital learning tools to loop playback or manipulate playback speed despite their availability for over two decades.
“We started this research planning to build a specific tool for ear learners, but then we realized we might be reinforcing a negative pattern of building tools without knowing what users actually want,” said Dan Brown, professor of Computer Science at Waterloo. “Then we got excited when we realized YouTube could be a helpful resource for that research process.” More