“Attention is all you need.”

This 2017 breakthrough idea transformed AI. The concept of self-attention became the foundation of today’s chatbots. Claude, Gemini, and ChatGPT are all large language models (LLMs), AI systems designed to focus on the matter at hand while filtering out distractions.

The results have been remarkable. From brainstorming recipes to generating code, apps, websites, and content, LLMs are being woven into our lives at breakneck speed.

But now, a City University of New York team and collaborators are asking: How closely does AI self-attention resemble human attention?

It’s not just academic curiosity. AI researchers have long looked to the brain for ideas to improve machine intelligence. In turn, AI models have offered new ways to investigate the brain. Comparing artificial and biological attention could inspire AI that concentrates more like us.

In their study, the team asked multiple chatbots to complete a classic psychology test of attention and cognitive control. Participants are shown the word for a color—such as “red”—written in either the same or a different color than the one the word describes. The challenge is to name the ink color while ignoring the word itself.

On short word lists, the chatbots performed at a high level. But as the tasks grew longer, their focus faltered. Instead of naming the ink color, they increasingly defaulted to reading the word. Under more demanding conditions—ones that also trip up people—their performance nearly collapsed.

The findings suggest today’s AI attention systems are “fundamentally limited,” wrote the authors. They go on to say that adding mechanisms similar to “those in biological attention is crucial for achieving artificial general intelligence.”

Attention, Two Ways

Doomscrolling. YouTube. Dinner plans. Family obligations. A barrage of notifications.

Life sometimes seems like everything, everywhere, all at once. Yet the brain can usually lock onto what matters most and push everything else into the background.

Far from a single, straightforward mechanism, attention emerges from multiple brain regions. According to attention network theory, three networks do most of the heavy lifting.

The alerting network keeps the brain ready for action. The orienting network selects which sights, sounds, smells, and sensations deserve attention. Finally, the executive control network resolves conflicts between competing streams of information, helping direct thoughts and actions toward a goal.

Together, these systems allocate the brain’s limited resources. Touch a hot stove, for example, and your brain immediately shifts attention to the burn over dinner. The food can wait; cooling your hand can’t.

AI works very differently.

Rather than processing language as complete sentences, LLMs break text into smaller units called “tokens.” Attention mechanisms then determine which tokens matter most for generating the next word, sentence, or response.

Self-attention is the key breakthrough behind modern chatbots. For each token, the model weighs and incorporates information from other tokens in a sequence, allowing it to track context across long stretches of text. This mechanism helps AI connect words and ideas, and underpins virtually all frontier LLMs today.

Researchers have since built on the concept. One approach, multi-head attention, runs several attention systems in parallel, with each “head” learning different patterns, such as grammar, syntax, or meaning. Another, cross attention, links information across different chunks of inputs and their outputs, making it especially useful for tasks such as translation and summarization.

But attention comes at a steep computational cost. To make models more efficient, researchers are also exploring sparse attention, which limits how many tokens a model considers at once. Another approach draws on information learned in the past to keep AI “focused.”

Despite the name, AI attention is ultimately a mathematical system. It helps determine what information is relevant in a specific context. But it lacks executive control, the network that keeps humans continuously focused on a goal despite distractions for long periods of time.

Color Blind

To test the limits of AI attention, the team pitted OpenAI’s GPT-4o and Anthropic’s Claude 3.5 Sonnet against the Stroop task.

Invented by John Ridley Stroop in 1935, the test measures attention and cognitive control by forcing participants to resolve conflicting information. The challenge is simple: Name the color of a word while ignoring what the word means. In a congruent trial, the word “blue” appears in blue ink. In an incongruent trial, “blue” might appear in red or green, creating a conflict between what the eyes see and what the brain reads.

Humans are consistently slowed down by this interference. Even with practice, the effect remains, suggesting it taps into fundamental mechanisms of executive control.

In the study, the researchers created word lists of varying lengths and difficulty. Some were entirely congruent. Others were fully incongruent. A third set mixed the two conditions.

At first, the AI models excelled. On five-word tests, GPT-4o was over 90 percent accurate across all conditions. But as the number of words increased, performance plummeted. On 40-word incongruent tests, the model’s accuracy fell to roughly 15 percent. Claude showed a similar decline. In mixed-condition tests, both models’ performance nearly collapsed to zero.

“The sharp decline in color-naming accuracy with increasing list length indicates that transformer-based attention mechanisms are vulnerable to scaling demands,” wrote the team.

Perhaps most intriguing, some models correctly recognized they were taking the Stroop test and could even explain its rules. But that apparent awareness did nothing to improve their scores. In other words, a “book smart” understanding of the task wasn’t enough to execute it well.

The study joins a growing effort to borrow psychological tests for research in machine cognition, especially when AI is challenged with complex, dynamic decision-making tasks. Theory of mind tests, for example, let researchers gauge whether a system can track others’ beliefs, emotions, and intentions. Personality tests are helping shape model behavior and reduce sycophancy. And some LLMs are readily solving emotional intelligence tests, which measure how well the algorithms recognize and respond to social cues.

According to the authors, the new results point to a missing ingredient in AI attention: A mechanism similar to the brain’s executive control network, which helps us stick to a task and adapt when priorities change.

Future AI systems could benefit from higher-level executive control that continuously tracks progress toward a goal, detects when attention has drifted, and pulls it back on course, if necessary.

Rather than simply weighing which tokens are most relevant in the moment, a more human-like form of attention could help AI stay focused during complex tasks, such as long conversations, multi-step reasoning problems, or high-stakes use in scientific research and drug discovery.

“The ultimate goal of AI research is to develop artificial general intelligence comparable to human abilities,” wrote the team. “AI systems, like humans, may need to master fundamental attention mechanisms…before achieving the generalized problem-solving abilities characteristic of mature executive functions.”

The post AI Collapses on a Classic Psychology Test. What It Reveals Could Stall Human-Level AI. appeared first on SingularityHub.

A handful of healthy senior citizens are about to trip on psilocybin—to see if the psychedelic protects aging brains.

Psilocybin, the active ingredient in magic mushrooms, is best known for its ties to 1960s counterculture. But now it may also herald a new genre of mental health treatment. From severe depression to post-traumatic stress disorder, studies have highlighted psychedelic drugs’ ability to reshape brain networks and relieve debilitating symptoms.

Most of these studies have focused on younger people with mental health conditions that don’t respond to standard treatment. The field’s success is prompting scientists to ask if psychedelics could also help healthy brains age better.

A team from the UC Berkeley Center for the Science of Psychedelics is about to find out. In a first-of-its-kind study focused on adults between the ages of 60 and 85, they’ll investigate how psilocybin affects perception, emotion, and memory using a battery of psychological tests.

Multiple scans before and after dosing will track changes in the brain. And detailed surveys will gauge broader shifts in well-being: Do participants feel more “in tune” with their emotions, feel less isolated, or experience a renewed sense of wonder about the world?

“What really excites me is that we’re focused on healthy older adults,”  said Tyler Toueg, who co-led the study’s design, in a press release. “Most clinical trials with older adults are focused on people who already have a diagnosis. We’re asking whether we can actually promote positive outcomes in older adults who are healthy.”

Called PLASTICITY, the trial could also open a rare window into how a psychedelic experience reshapes healthy brain networks. And because the drug alters our sense of self, psilocybin could help researchers probe the ways in which the brain constructs reality.

“I’m very interested in psilocybin as a potential mental health treatment, but I’m also interested in it as a way to shed light on these central mysteries in neuroscience and psychology,” said study designer Michael Silver.

Chasing the White Rabbit

Psychedelic research was highly restricted for decades. But advocates, including the non-profit Multidisciplinary Association for Psychedelic Studies, have steadily pushed to reopen the field, arguing that these drugs might keep mental health symptoms at bay.

Early results helped usher psychedelics into the mainstream. In 2023, a randomized, placebo-controlled trial found that a single dose of psilocybin, paired with therapy, eased depression. Oregon later approved supervised psilocybin therapy—though the drug remains federally illegal in the US—and Australia became the first country to greenlight it for depression and post-traumatic stress disorder. More recently, two late-stage studies reported strong effects in severe depression, potentially paving the way for FDA approval.

Scientists still don’t fully understand how psilocybin works in the brain. But there are hints. The drug appears to rapidly reorganize the connections between brain cells, particularly in the hippocampus, a region of the brain central to learning, memory, and navigation.

Neurons constantly change their connections in a process called plasticity that encodes experiences into neural networks, allowing the brain to process information, learn, and lock in memories. In youth, these connections are flexible and expansive. But with age and in conditions like depression, the brain’s flexibility wanes.

The birth of new neurons, or neurogenesis, also contributes to plasticity. Neurogenesis only occurs in two brain regions, one of which is the hippocampus. Although whether it actually takes place in humans is controversial, it is strongly linked to learning, memory, and emotion, and it declines in both psychiatric disorders and aging.

Psilocybin may reset brain plasticity to a more youthful state.

In rats modeling depression, for example, a study showed the drug shifted dark moods into behavior that was more exploratory and engaging. Where traditional antidepressants, like Prozac, tend to blunt symptoms, psilocybin seems to overwrite entrenched negative patterns. This suggests deeper circuit-level changes.

In another study, the drug reopened a critical window for learning in mice. During adolescence, the brain is highly plastic, but it begins to stiffen in adulthood. Psilocybin temporarily restored malleability and changed the mice’s social behavior. In some brain regions, the drug increased sensitivity to oxytocin, the so-called “love hormone.” The authors suggested the drug induced a state called metaplasticity, in which neurons are more likely to respond to oxytocin and other regulators to rewire, form new connections, and grow their branches.

Psilocybin’s effects may extend beyond the brain. Depression, chronic stress, and the immune system are tightly linked. In a third study, researchers identified a brain-spleen connection that drives fear and anxiety. Psilocybin suppressed inflammatory immune cells associated with brain inflammation and dampened anxiety-like behavior in stressed mice, even under threat.

Both effects—dialing up plasticity and lowering inflammation—make psilocybin an intriguing way to potentially counter changes in the aging brain.

“We know that with age, we lose synaptic connections, especially in certain brain regions like the hippocampus and prefrontal cortex,” said Toueg. “There’s a lot of overlap between the mental states that psychedelics influence and those associated with successful aging.”

Tomorrow Never Knows

The PLASTICITY trial is designed to test whether psilocybin can produce lasting changes in neuroplasticity in healthy adults aged 60 to 85. Participants will first undergo assessments of cognition, visual perception, and brain structure using advanced MRI techniques.

Diffusion MRI scans will focus on the hippocampus to capture microscopic changes in its structure. Functional MRI will focus on brain activity as participants perform learning and memory tasks, offering a dynamic view of how activity shifts after dosing.

The study will also see if psilocybin increases vagus nerve activity, which has been linked to better stress recovery. Participants will complete detailed surveys about the experience ranging from emotional responses, like wonder, to potential shifts in outlook and social cognition.

“Things like depression, anxiety, stress and rumination are all associated with worse aging outcomes,” said Toueg. “Things like having purpose in life, emotional regulation, and awe are all associated with more successful aging.”

The trial began enrolling participants in November last year. Two volunteers have already completed the tests, and the team aims to dose 20 people by the end of 2026.

Although psilocybin trials are now widespread, older adults remain underrepresented. One estimate suggests just 1.4 percent of participants are 65 or older, despite potentially being among those most likely to benefit from interventions that enhance plasticity.

“This study allows us to directly test whether the promising findings from animal models translate to older humans and to generate data that will inform future research on aging, cognition, and mental health,” said Silver.

Toueg agrees. “I think that no matter what we find, this study will have implications for how we think about intervening in the aging brain,” he said.

The post Can Psychedelics Reboot Aging Brains? We’re About to Find Out appeared first on SingularityHub.

Scientists at Columbia University have used a precise gene-editing tool, base editing, to make changes in three disease-linked genes in early-stage human embryos. The goal wasn’t to create pregnancies, but to test the safety and limits of rewriting DNA at the very early stages of life.

The paper, not yet peer reviewed, sparked immediate controversy. Some researchers hailed it as a technical milestone that could one day prevent devastating inherited diseases before birth. Others warned it edges society closer to the prospect of “designer babies”—an idea bioethicists have argued is akin to modern eugenics.

The debate is hardly hypothetical. The work has already attracted commercial interest. New York-based Nucleus Genomics, which screens in vitro fertilization (IVF) embryos for serious genetic disorders, has also developed predictive models for complex traits such as intelligence. The company plans to sponsor future research by study leader Dieter Egli and team.

Critics worry that even experimental advances could fuel demand from wealthy patients while encouraging companies to develop and market embryo-editing technologies, despite unresolved ethical and safety concerns.

Egli argues the findings should be public precisely because these debates are no longer academic curiosity. He has repeatedly called for scientists, regulators, and the public to weigh the pros and cons of editing human embryos. As for clinical use today, his position is unequivocal: “You can’t use it. It’s as clear as day and night,” he told Nature.

Conceptual Shift

Why edit embryos at all?

Cells in an early embryo eventually give rise to every tissue in the body. Correct a harmful mutation at the start of development, and the fix could, in theory, propagate throughout a child’s entire body—and even be passed on to future generations.

The strategy could help in genetic disorders that hamper fetal development or trigger diseases in newborns. For some developmental and metabolic conditions, intervention after birth may already be too late. Even when treatment is possible, gene editors must be able to target various organs, which is an ongoing challenge.

In various efforts, scientists have already repaired disease-causing mutations in mouse embryos and fetuses, including those linked to blood disorders. But mice aren’t humans. Early embryos from the two species repair DNA damage in fundamentally different ways, making it tough to gauge whether a strategy that works in mice will succeed, or prove safe, in people. That uncertainty has fueled interest in testing gene-editing tools directly in human embryos.

Not everyone is on board. International scientific groups have repeatedly called for a temporary ban on editing human embryos, and the practice is illegal in several countries.

That didn’t stop Chinese scientist He Jiankui. In 2018, he announced the birth of gene-edited babies after using a tool called CRISPR-Cas9, claiming the changes would protect them against HIV infection. Global outrage ensued.

By then, years of research had already highlighted CRISPR’s risk. The tool cuts both strands of DNA and relies on the body’s repair machinery to stitch them back together. But the process can go awry, introducing unintended mutations, deleting large chunks of DNA, or altering the wrong locations on the DNA strands altogether. He’s reckless experiment resulted in three years of imprisonment, although he still defends the work.

Subsequent studies only deepened concerns. In some cases, CRISPR editing in human embryos caused extensive genetic damage. In one study,  it completely destroyed the chromosome that housed the target gene.

An Imperfect Upgrade

The new study tested a next-generation gene editor designed to overcome some of CRISPR’s biggest shortcomings.

Egli and team used an approach called base editing, which rewrites individual DNA letters. Unlike CRISPR, base editing only nicks the DNA strands and is generally thought to be more precise. The technology hit a major milestone last year when it helped cure a baby with a potentially fatal genetic disorder, and earlier lab studies hinted it could also succeed in human embryos.

Working with early-stage embryos, the team edited three genes with the potential to cause illness. In each case, they converted the genetic letter A to G at precise locations. One of the genes, PCSK9, regulates “bad” cholesterol levels. Mutations are associated with a high risk of heart problems. The team’s edit was designed to switch off the gene, mirroring strategies already being explored in adults.

The other two targets, HBG1 and HBG2, control production of fetal hemoglobin, an oxygen-carrying protein. The edits made here reflected a natural protective variant that could lessen symptoms in blood disorders, such as sickle cell disease and beta thalassemia.

The team found no signs of widespread DNA damage, suggesting the tool is more precise than CRISPR. But it wasn’t perfect. Many embryos emerged as so-called genetic mosaics, with some cells carrying the intended edit and others retaining their original genetic blueprint.

That’s a huge problem. As an embryo develops, unedited cells could outcompete edited ones, leaving the disease-causing mutation largely intact. In some embryos, edited cells stopped dividing altogether.

And a lack of obvious chromosome damage doesn’t guarantee safety. The edits could still trigger harmful effects that aren’t noticeable until after birth—when it’s already too late to reverse them.

Calls for Scrutiny

Egli stresses that embryo editing is still far from being ready for the clinic. “These base editors—they can have damaging effects on the embryo. So why would you use it if you don’t fully understand that?” he told Nature.

His team is now working to reduce mosaicism and plans to test the technology in embryos that have developed to roughly 100 cells. This is when fertility clinics typically evaluate and freeze embryos.

Speaking to The New York Times, fertility expert Paula Amato at Oregon Health & Science University, who was not involved in the work, called the strategy “promising.” Genomics researcher Greg Neely at the University of Sydney in Australia also praised the work: “This will go down in history in a positive way—less reckless, more careful and ethical than previous attempts.”

Others remain deeply skeptical. Critics argue that embryo editing permanently alters the genetic inheritance of future generations, who have no say in the decision. The study’s ties to Nucleus Genomics also raised eyebrows. The company previously drew controversy for developing genetic predictions for traits such as intelligence and height and for its slogan “have your best baby.”

To Kian Sadeghi, CEO and cofounder of Nucleus, embryo editing extends that vision. The technology could help couples carrying mutations who struggle to produce enough unaffected embryos for selection during IVF.

Fyodor Urnov at the University of California, Berkeley, who was not involved in the study, isn’t convinced. IVF clinics already screen embryos for many inherited disorders without altering their DNA. Given the risks, selecting an unaffected embryo is often a safer option than rewriting its genome.

“In practical terms, therefore, this preprint will solely impact the rapidly growing movement of embryo editors for purposes of ‘baby improvement’,” he said.

That movement, once taboo, is gaining steam. Yet the traits most often cited by proponents—height, intelligence, emotional regulation—are shaped by hundreds or even thousands of genes, which scientists still don’t fully understand. Such enhancements are far beyond the reach of today’s technology. Every additional edit also increases the chance of unintended consequences.

For Egli, that’s precisely why the research should be discussed openly. “Research is necessary to provide information to discourage the wrong use of a technology,” he said.

The post Precise Gene Editing in Early Human Embryos Reignites the ‘Designer Baby’ Debate appeared first on SingularityHub.

AI is becoming more powerful, and mysterious.

Despite years of work on “explainable AI,” today’s most advanced systems remain black boxes for the most part. Scientists can observe what they do but cannot fully explain how they arrive at their conclusions or predict when they’ll fail.

As large language models (LLMs), the algorithmic engines behind popular chatbots, permeate society, researchers are warning that the window for understanding AI “minds” is rapidly closing even as the technology’s influence expands.

Last week, Eric Horvitz, chief scientific officer at Microsoft, and Robert West at EPFL in Switzerland outlined the dangers of putting AI interpretability on the back burner. They call for new AI benchmarks and better tools for unpicking machine minds.

The challenge resembles efforts to understand our own minds. Some researchers have already taken a neuroscience-inspired approach, mapping AI’s internal networks to concepts, goals, and reasoning. Others borrow from psychology, treating AI as a participant of behavioral studies.

The stakes are rising. AI tools already shape how people search for information, make decisions, and form judgments. Their answers influence everyday users and the researchers who build them.

As AI capabilities grow, our understanding of them could fall behind. “Preserving human agency must therefore remain a central goal,” the authors write.

The Black Box Conundrum

LLMs are built on artificial neural networks (specifically, a design called the transformer). Inspired loosely by the brain, these networks connect vast numbers of artificial neurons into intricate architectures. The basic idea is straightforward. Data enters the network and passes through layers of computations, which transform it into an output like text or code.

At first, that output is often wrong. But with feedback and repeated training, the network adjusts the strengths of connections between neurons and gradually improves. It learns.

After initial training, engineers turn to reinforcement learning, where algorithms improve through trial and error and further hone their responses. Another method, inspired by how the brain etches memories during sleep, reduces the tendency to forget old knowledge while learning new tasks. And self-attention, the key innovation behind transformers, allows AI to selectively focus on various words, images, sounds, or video frames at different moments, boosting efficiency and performance. Today, attention underpins nearly every major AI system.

Yet the inner workings of finished algorithms remain hidden.

Early efforts to crack open AI’s black box examined how artificial neurons responded to images, revealing that neural networks build increasingly more sophisticated “ideas” of the world. Google Brain borrowed methods from cognitive psychology to study AI behavior, while others investigated whether LLMs could mimic aspects of “theory of mind”—the ability to infer what others are thinking and feeling.

These studies laid the foundation for a popular method called mechanistic interpretability. Anthropic, creator of Claude, is leading the field. Company researchers have linked patterns of algorithmic activity to specific concepts and reverse engineered parts of neural networks to expose how internal computations shape responses.

Other tech giants are joining the cause. OpenAI is training algorithms that work in more explainable steps and building reasoning models that pause, “think,” and justify their conclusions in plain language. DeepMind is building microscope-like tools for neural networks, helping researchers peer into their decision-making process. And Microsoft has released new tools aimed at responsible use of AI.

Understanding AI, the authors write, does not require tracing every line of code or every neural-network parameter. Just as neuroscience, psychology, and sociology offer different windows into human behavior, AI can be studied at multiple levels, from how individual circuits work to observing behavior in real-world scenarios.

The challenge is that AI capabilities may be advancing faster than our ability to explain them. And some researchers believe time is running out.

Race Against the Machine

Three trends are making AI more opaque.

The first is how we evaluate AI. Increasingly, LLMs we being used to train, benchmark, and improve other models. AI “judges” now score metrics like helpfulness, rank competing outputs, detect hallucinations, and assess new releases. In a system known as constitutional AI, for example, algorithms critique their own responses using reinforcement learning and generate explanations for their reasoning. Other researchers have proposed AI debate frameworks, where multiple models challenge each another’s conclusions before a human has the last say. Researchers are also exploring automated interpretability tools. Like digital neuroscientists, AI systems are used to analyze each other—describing neurons, circuits, and behavioral patterns—to explain increasingly complex models.

Using AI to solve an AI-induced problem introduces a paradox. If AI-generated explanations become too complex for humans to verify, opacity compounds.

A second trend is the rise of AI societies. Networks of interacting AI agents are becoming more common, particularly in complex tasks such as scientific research and drug discovery. Yet as they become more sophisticated, their communication could drift from human language and reasoning, making them harder to interpret.

Studying their interactions with methods adapted from sociology could unveil unexpected norms, hidden rules, and collective behavior. The authors argue that training in the future should not only reward effective collaboration among AI agents, but also ensure humans can understand their communication.

The last trend already permeates our lives. ChatGPT, Claude, Gemini, and other LLMs listen to our woes, offer recipes, and code websites. But they also learn about humanity. Through training data and interactions, they glimpse how people think, reason, and feel. In turn, they capture core aspects of life, such as fear, anxiety, happiness, and the need for social belonging.

To be clear, the systems don’t have intentions. They’re not examining us. But even as we struggle to understand them, AI systems are building more sophisticated models of who we are.

“A striking asymmetry follows: While human understanding of AI declines, AI understanding of humans deepens, producing new forms of behavioral opacity,” the authors write.

But complacency is perhaps even more insidious. AI assistants are often optimized to be agreeable, helpful, and reassuring. Studies have found that people generally prefer AI agents that support their opinions and decisions. As AI is woven into everyday life, curiosity and skepticism may gradually give way to trust. They work. Why question how?

The authors don’t have a solution for the long-standing problem. Instead, they call for better benchmarks to measure AI capabilities and stronger evaluation methods. And while open-source projects and crosstalk between commercial companies and academia are now frequent, they say we need lasting norms of responsible disclosure. Mechanistic interpretability and AI “psychology” could build on each other.

“The goal is not just more capable AI, but AI that is more intelligible, accountable, and aligned with human aims,” they write.

The post AI Is Advancing Faster Than Our Ability to Understand It, Researchers Warn appeared first on SingularityHub.

For decades, a handful of molecular switches has haunted the nightmares of cancer researchers. The switches trigger runaway tumor growth and cause the disease to spread across the body in multiple cancers. In theory, this makes them perfect treatment targets. Blocking even one could lead to drugs that are effective against a variety of cancers.

But despite considerable efforts, these switches—all of which are proteins—have escaped our most advanced cancer treatments, earning them the term “undruggable.” This is largely due to a shared trait: They all have smooth surfaces, making it difficult for drugs to interact with them.

But maybe not for much longer.

Researchers recently reported promising results for a new medication targeting a family of undruggable proteins in a clinical trial for advanced pancreatic cancer. The drug, daraxonrasib, nearly doubled survival time compared to chemotherapy, with fewer side effects. It’s not a total cure. But the treatment gives patients precious time, adding roughly 13 months after diagnosis. Patients also reported less pain and better quality of life.

Daraxonrasib is the latest in a new generation of drugs aimed at undruggable proteins. And AI-based tools are now poised to further accelerate progress in the field.

RAS Attack

The RAS family was the first group of oncogenes—or genes that drive cancer—ever discovered. The genes became a major focus in 1982 when several teams independently showed the mutation of a single DNA letter could transform RAS genes into a potent cancer trigger.

The proteins RAS genes encode are like spring-loaded molecular switches that relay signals from a cell’s surroundings. When proteins called growth factors latch onto a cell, RAS switches flip on to promote cell growth and survival, while built-in safeguards quickly turn them off again.

Cancerous mutations break this cycle. The switches get stuck in the “on” position, continuously instructing cells to grow and divide. This is, of course, a hallmark of cancer.

An ideal drug would simply switch RAS off. But most drugs are like rock climbers. They need grooves, pockets, or bumps on a protein to grab onto. Similar to a smooth rock face, RAS offers few such features. Making matters worse, different mutations subtly reshape the protein, so it’s tough to build a one-size-fits-all inhibitor.

The first RAS drug wasn’t approved in the US until 2021, nearly four decades after discovering the genes’ role in cancer. Even then, the drug targeted just one family member of three, limiting its reach to a relatively small group of patients. Many eventually developed resistance.

That’s why daraxonrasib turned heads. Developed by Revolution Medicines in Redwood City, California, the drugs switches off all three RAS family members. Rather than trying to grip the slippery proteins directly, it binds to a partner molecule that helps RAS proteins fold into their final 3D shapes. In this way, the drug hitches a ride on active RAS and shuts the proteins down.

The workaround paid off. The new study enrolled 500 people worldwide with advanced pancreatic cancer. All participants had already tried cancer therapies with limited success. On average, patients receiving daraxonrasib lived 13.2 months and spent most of that time with limited pain. The most common discomfort was a rash. Those receiving chemotherapy fared worse, living roughly 6.6 months and experienced more severe side effects.

The results don’t rival the dramatic success of CAR T cell therapies in blood cancer. In CAR T, caregivers engineer a patient’s own immune cells to recognize and attack tumors, sometimes producing long-lasting remission after a single infusion.

But the findings have energized the field. If approved, a daily daraxonrasib pill would likely be far more affordable and easier to administer than a personalized cell therapy. And because RAS mutations fuel many solid cancers—which CAR T still struggles to control—the drug could offer a new defense against deadly cancers that are largely beyond cell therapy’s reach. Combining daraxonrasib with earlier-generation RAS inhibitors may further boost its effects.

The Genome Guardian

Daraxonrasib didn’t appear overnight. Scientists used a crystallized snapshot of its target protein as a molecular blueprint. Years of medicinal chemistry followed, with scientists repeatedly tweaking candidate compounds to boost potency, improve selectivity, and minimize toxicity.

AI could dramatically accelerate similar efforts against other undruggable cancer targets. Among the most coveted is p53, often called the “guardian of the genome” for its dizzying array of roles. The protein orchestrates the activity of over 300 genes involved in DNA repair, metabolism, cell death, and inflammation, making it one of the cell’s most important defense systems.

Since its discovery in 1979, p53 has been both a holy grail and a headache for cancer researchers. Mutations in the gene are common in multiple cancers. But like RAS, the protein is flat and smooth. Some mutations destabilize its structure; others turn it into misfolded clumps. A universal p53 drug has remained elusive.

Some researchers are trying to restore the protein. In a small trial earlier this year, they tested a drug that restabilizes a common mutant form of p53. Within 21 days, tumors shrank roughly 20 percent in patients with ovarian, breast, and several other solid cancers.

Other researchers aim to selectively kill cells carrying the mutation. Using AI, a team at Baylor College of Medicine screened nearly 10 million compounds that cause mutated p53 cells to self-destruct, while sparing healthy cells. The search uncovered 83 chemically distinct candidates. One called H3 dramatically suppressed tumor growth in mice.

“These results highlight the potential use of AI-powered drug screening to investigate individual p53 mutants in the future,” they wrote. Although the approach is early-stage and only focused on one mutation, the team is hopeful it can be extended to other cancerous mutations.

Most Wanted

MYC is another formerly undruggable protein that could now be vulnerable. Roughly 70 percent of cancers have abnormal MYC activity. Normally, the protein is a master regulator of growth, directing cells to manufacture proteins, replicate DNA, absorb nutrients, and divide when needed.

Cancer finds many ways to hijack the system and keep cells in a state of runaway growth. MYC gene mutations aren’t just single-letter swaps. Sometimes the gene duplicates or is rearranged across the genome, churning out excessive amounts of the protein it encodes. This genetic diversity makes approaches using gene therapy difficult. And again, like RAS, the MYC protein’s smooth, featureless surface lacks stable anchors for drugs.

An emerging strategy is to disrupt MYC’s interaction with other proteins that it needs to function. A designer protein blocking MYC activity, for example, recently showed promise in a small trial against solid cancers. Other teams are using AI to identify drugs that limit MYC’s ability to fix damaged DNA in tumors, kneecapping their ability to divide. Meanwhile, biotechnology companies are deploying AI to map out MYC’s structure and molecular interactions in search of new ways to shut the protein down.

Daraxonrasib’s success shows that undruggable proteins aren’t untouchable. There’s a lot more work ahead to prove other similar drugs can work too. But scientists are increasingly leaning into AI during all stages of drug development to speed up the process. Maybe, one day, “undruggable” will disappear from our vocabulary altogether.

The post After Decades of Failure, ‘Undruggable’ Cancers Begin to Give Way appeared first on SingularityHub.

Huntington’s disease is tragically predictable. An inherited genetic mutation causes neurons to make distorted, sticky proteins. These proteins clump together and gradually overwhelm brain cells. The brain loses its ability to learn, remember, and make decisions.

This story is dogma in neuroscience. But decades of research and drugs targeting the clumps have had little success. Scientists are now wondering: Is there more to the story? In a twist, a team from the Hebrew University of Jerusalem and collaborators found that protein clumps may be a neuron’s first line of defense against damage.

The misfolded or malfunctioning proteins are quarantined inside bubbly hubs called “inclusion bodies.” Often considered detrimental to cell health, disrupting their formation unexpectedly led to cells becoming more sensitive to stressors often seen in neurodegenerative diseases.

Physical separation played just one part. Inclusion bodies also changed the activity of genes involved in neuroinflammation—even in the absence of immune cells. Scouting the genetic landscape of cells derived from patients with severe Huntington’s disease, the team homed in on a “master regulator” gene, ATF3, that orchestrates immune responses. Removing the gene lessened inclusion bodies’ protective effects against damage in cultured cells.

To be clear, the findings are only for a cell model of Huntington’s disease in a petri dish. And inclusion bodies could be a double-edged sword: protective in the beginning and detrimental later on. Still, acknowledging them as a more complicated villain could better inform strategies for disorders that take over our minds like Huntington’s.

“Our results reveal…that these structures are not merely byproducts of disease, but a central factor in the cell’s ability to mount a protective response against stress,” said study author Eran Meshorer in a press release.

The Problem With PolyQ

It’s long been believed that protein clumps in the brain gradually erode cognition. Whether they’re the main driver of neurodegenerative disorders is still debated, but their presence accelerates brain cell injury, causing neurons to wither away.

Alzheimer’s disease, for example, is associated with two sets of protein clumps. One lives inside neurons (tau) and another gunks up the space between cells (amyloid). Decades of research aimed at removing amyloid clumps have met with minimal success, earning these doomed efforts the notorious nickname “graveyard of dreams.” Despite their struggles, the FDA recently approved two major drugs that remove amyloid clumps and modestly slow cognitive decline, though the approval has been controversial due to doubts about safety.

Other untreatable neurodegenerative disorders also fall into this category. Clumps formed in Parkinson’s disease erode the brain’s ability to control movement, emotion, and even the perception of time. Lou Gehrig’s disease, or ALS, produces inclusion bodies inside motor neurons, leading to muscle weakness and trouble swallowing. The disease eventually robs people of speech and motion.

These diseases often have multiple genetic and environmental triggers. Huntington’s, in contrast, is entirely genetic. The condition stems from the genome over-copying parts of the huntingtin gene (HTT), which normally makes a key protein also called huntingtin.

Normally, cells use the protein’s large, stackable structure to build highways that transport all sorts of biological cargo, from molecules to organelles. The protein also plays an essential role during early brain development and neural wiring in adulthood.

But a mutant form of the HTT gene can wreak havoc. A common mutation, called polyQ expansion, produces unwieldy, misfolded proteins. Nearly 30 years ago, researchers found that these errant proteins aggregate inside parts of the cell. The clumps, or inclusion bodies, were widely thought to be detrimental. Some act like sticky tape that captures healthy proteins, such as those involved in gene expression, and torpedoes cellular health.

But telltale signs in cultured rat brain cells suggest a more nuanced story: Inclusion bodies could also be protective, sequestering mutant proteins as an early form of protection.

A Tale of Two

The common factor in diseases featuring polyQ mutation is repetition. Mutated genes have long, duplicated sequences of the DNA letters cytosine, adenosine, and guanine (CAG). More CAG repeats in the genome translates into earlier disease onset.

We all have this DNA triplet in our HTT gene. But more than 39 repeats results in longer, toxic huntingtin proteins. Severe cases of Huntington’s can feature over 100 CAG repeats, transforming the usually free-floating protein workers into sticky, dysfunctional layabouts.

In the new study, the researchers first established a baseline. They used the gene editing tool CRISPR-Cas9 to reduce CAG repeats in cells derived from Huntington’s patients—which carried over 180 copies—to near normal levels.

They then tagged the cells with a fluorescent marker that causes huntingtin proteins to glow bright green under the microscope. This let the team track protein aggregation in real time. Though they shared the same genetics, some cells formed inclusion bodies; others didn’t.

The team next challenged them with a chemical known to cause cellular stress. Those that formed clumps survived far more regularly than those that didn’t. It was a “striking difference,” the authors wrote. “Once a mutant PolyQ protein is expressed, the formation of IBs [inclusion bodies] protect[s] the cells rather than inflict[s] harm, at least short-term.”

Inflammation seems to be key. Although grown side-by-side, a genetic screen revealed cells with inclusion bodies were especially abundant in a gene called ATF3, which is known to regulate inflammation. Getting rid of the gene wiped out the neurons’ ability to form inclusion bodies, making them more vulnerable.

“Our results reveal a previously unknown role for ATF3 in orchestrating the formation of inclusion bodies in human neurons,” said Meshorer.

These are very early results. An immune molecule bridges ATF3 and inflammation and is associated with Huntington’s disease. Its levels are higher in patients with the condition. Increasing ATF3 activity could amp up the number of protective inclusion bodies and give neurons a fighting chance.

The findings suggest inclusion bodies gather free-floating mutant proteins into clumps to protect neurons and reduce brain damage—at least at the beginning of the disease. However, lab experiments rarely translate to treatments. How fast inclusion bodies form and when they begin to stress cells remains to be seen. Meanwhile, a gene therapy for Huntington’s is underway, and promising results in a small trial suggest an alternative path for treatment.

Still, the study challenges the idea that protein clumps are always detrimental. If replicated in other neurodegenerative diseases such as Alzheimer’s or ALS and if we can learn how long protection lasts, the results could pave the way for better-timed treatment that works with the body’s protection, not against it.

The post Toxic Clumps in Huntington’s Disease May Protect the Brain Too appeared first on SingularityHub.

Lithium mining is like a modern gold rush. The element is the main ingredient in batteries powering smartphones, electric cars, and even AI. Global demand is surging. Increased production could guide the world toward a more sustainable energy future.

But ironically, current extraction methods offset some of those gains. Lithium mining involves separating the element from brines using toxic chemicals, a process that also pumps out carbon dioxide. This, alongside enormous water and energy costs—due to high temperature requirements—has confined mining to a handful of countries.

To address these drawbacks, scientists at the Massachusetts Institute of Technology have now developed a low-cost, low-temperature, greener process relying on an abundant resource: Hard rock. Although rocks containing lithium cover large parts of the US, Europe, and Africa, extracting it from them is challenging.

While renovating his bathroom, study author Yet-Ming Chiang realized a chemical in glass etching cream—which makes glass translucent—could eat away at lithium-rich rocks. His team then designed a recyclable process to extract lithium as well as two ingredients used to make greener cement and other materials.

“You’ve heard of nose-to-tail eating?” said Chiang in a press release. “We refer to this as nose-to-tail mining.”

Unlike previous methods, the process runs at temperatures below the boiling point of water. All liquid chemicals are almost recyclable and can be reused in multiple rounds of extraction.

“This could establish a low-carbon alternative to hard rock refining, addressing both the surging demand for lithium and the carbon footprint that undermines the sustainability of the energy transition that lithium is meant to enable,” wrote Gang San Lee and Karthish Manthiram at the California Institute of Technology, who were not involved in the study.

A Rock and a Hard Place

The Earth’s crust teems with lithium. Getting it out is the hard part.

Currently, many mining operations rely on brine that naturally leaches lithium over millennia. Later steps purify the lithium into a battery-ready product. The process relies on large evaporation pools and is limited to a few countries, making the resource scarce.

Lithium could, alternatively, be harvested from solid rocks. One ore, spodumene, is packed with lithium, roughly 1.5 percent by weight. But liberating it has been a tough nut to crack.

Traditionally, miners crush rocks and remove chunks that don’t contain lithium. The rocks are then blasted at temperatures as high as 1,100 degrees Celsius (2,012 degrees Fahrenheit) and showered in a cocktail of dangerous chemicals. The process spews liquid waste into the environment and releases 20 tons of carbon for each ton of lithium.

Researchers are working on more temperate methods.

One of these is called ball milling. Ore is rotated in a container filled with hard balls that mechanically grind the stone into a fine power. It’s like using a mortar and pestle instead of a blender. But the process takes longer, and lithium is lost along the way, resulting in lower yields. Another method, called electrochemical leaching, refines the ore at room temperature. But researchers have had mixed success with the process, and it’s tough to scale up. It also produces in a lot of waste rock that could, in theory, be harvested for other uses instead being discarded.

Triple Threat

The new method popped into Chiang’s mind as he was brainstorming ways to break apart spodumene, a lithium-rich ore with high amounts of silica—the main ingredient in glass.

Dissolving silica to get to lithium requires hydrofluoric acid, a highly toxic chemical. But glass etching cream also eats away at silica with ammonium fluoride. Tubes of the mild acid are available in home improvement stores, and it works at room temperature. Why not give it a try?

By mixing ammonium fluoride with water, the team showed they could completely dissolve spodumene at temperatures below 100 degrees Celsius without releasing toxic fumes. They only needed to continuously stir the ore in a simple plastic tank. The process yielded several types of lithium salt with 99 percent purity. In early experiments, extraction took several days, but the team has since cut the time to under 12 hours.

“Dissolving silica is the hard part in mining,” said study author Benjamin Mowbray. “The next question was how do we apply it to impactful mineral processing problems?”

Along with lithium, spodumene is jam-packed with two usually discarded ingredients: Alumina, which after smelting makes aluminum, and silica, which can be directly used as a sustainable ingredient in greener cement. The new process can separate out both materials, and the team vetted the resulting products, including strength testing cubes of fabricated cement.

“First our goal was to produce these products, then there were additional steps of characterizing their purity and properties and making sure our products met the specifications for target markets,” said Mowbray.

“If any product didn’t meet the target specs, you’d end up with a waste stream.”

With a few chemical tweaks, the team showed the acid could be regenerated and reused at least five times. The team successfully processed 17 spodumene ores sourced from around the world, suggesting the method could be broadly applicable.

They’ve also spun the work into a startup, Rock Zero, and aim to scale it. If the acid can be recycled with near-perfect efficiency, the team estimates the process would cut costs over 40 percent compared to conventional hard-rock extraction, making it competitive with brine operations.

Its simplicity could also reshape where lithium gets produced. In 2024, roughly 74 percent of global lithium output came from just three countries: China, Australia, and Chile. By eliminating the need for extreme heat and massive waste-treatment plants, the process could be easier to implement, especially in countries rich in spodumene but lacking the capital for infrastructure.

That opens the door to a network of smaller refineries built closer to the mines themselves, reducing transportation costs and supply-chain bottlenecks. Because the process is also far less energy intensive, it could be powered by solar and wind, further shrinking its environmental impact.

The technology could also be adapted to recover other valuable metals hidden inside mineral ores. One candidate is beryllium, a lightweight but extremely stiff and stable metal used in satellites and the James Webb Space Telescope’s mirrors. Current manufacturing processes often generate toxic dust and fumes linked to serious lung inflammation. A cleaner extraction route could make it safer and cheaper to produce.  

As for Rock Zero, going up against established lithium giants is like David and Goliath. They’ll also have to contend with global market volatility and increasing competitiveness of sodium-ion batteries and other alternative battery chemistries.

But the team is unfazed. “We believe this approach is the lowest-energy, lowest-cost way of getting lithium not only out of hard rock, but period,” said Chiang. “That’s what’s motivating us to scale this.”

The post Three Countries Own the Lithium Market. An MIT Startup Wants to Break Their Grip. appeared first on SingularityHub.

There’s truth to the old adage, “Age is just a number.” People of the same age differ vastly in health and mental capabilities. One 80-year-old may be vibe coding with Claude, while another is gradually forgetting familiar faces and memories.

To better gauge this difference, scientists have been developing “clocks” that measure biological age. Rather than the number of candles on a birthday cake, these tools capture health at the cellular level and are remarkably accurate at estimating disease risk and even life expectancy. But how they work is hard to explain.

Now Harvard scientists and collaborators have released a powerful and more interpretable clock. Using the gene activity of thousands of individuals and animals, the clock predicts biological age in rodents, monkeys, and humans, including how many years they have left.

The analysis involved over 11,000 gene activity profiles across four species, highlighted shared mechanisms during aging, and responded to known anti-aging interventions—such as parabiosis, during which aging animals receive blood from a young donor.

Although the clock isn’t ready for clinical use, it is a boon to scientists working to slow or even reverse the unstoppable progression of time. It “could help researchers to pinpoint which processes are modulated by interventions or diseases,” wrote João Pedro de Magalhães at the University of Birmingham, who was not involved in the work.

Tick, Tock

Biological clocks come in a variety of flavors.

Most rely on AI to make sense of information held in large databases of people. One of these, for example, uses blood proteins related to brain aging to reflect cognition and its decline better than chronological age. Another type, metabolomic age clocks, sorts through protein and fatty acid building blocks to estimate biological age. These clocks correlate well with risk of inflammation, chronic disease, and frailty (where the body struggles to recover from a mild infection or minor fall). More recent multi-omics clocks combine blood measures, metabolism, gene activity, and clinical data for a comprehensive bird’s-eye view of biological age.

But epigenetic clocks remain the field’s defining breakthrough.

As we age, chemical tags accumulate on DNA, switching genes on or off. The pattern of these tags shifts over time and is shaped by everyday life—diet, exercise, stress, sleep quality. Studies have found that the age gaps between biological and lived years measured by the well-known Horvath epigenetic clock, which relies on DNA methylation, were associated with the risk of various types of diseases. Later versions of the Horvath clock could predict maximum lifespan. And other groups have developed “pan-mammalian” epigenetic clocks that work across species.

“One drawback of epigenetic clocks, however, is their limited interpretability,” wrote Magalhães. “The mechanisms that underpin age-related methylation changes are still debated.”

Clocking In

In the new study, the team measured aging by looking at gene activity, or transcriptomics. Transcriptome profiles capture which genes are switched on at any given moment.

Previous studies have linked the aging transcriptome to chronic inflammation, faltering mitochondria, and the gradual breakdown of the extracellular matrix, the molecular scaffolding that supports tissues and organs. With age, these systems go awry.

“Because the signatures reflect changes in the activity of specific genes, transcriptomic biomarkers are more interpretable than are epigenetic ones,” wrote Magalhães. The tradeoff is that gene activity is far more dynamic than DNA methylation, the epigenetic signature used in the Horvath clock. A transcriptome can shift in response to stress, illness, exercise, or even the time of day, making it a less reliable measure of aging.

To make the new clock, the team assembled over 11,000 transcriptomes, heavily relying on data from the Interventions Testing Program, a giant effort to study longevity treatments in mice. The dataset included mice exposed to genetic tweaks, drugs, and dietary therapies known to affect aging and lifespan. The team also added more than 2,600 samples from monkeys, several hundred from rats, and over 4,000 from humans to deliver a cross-species view of aging.

They then built multiple transcriptome clocks that estimated age and mortality risk. To validate the clocks, they turned to an independent dataset that included rodent models of accelerated aging, Alzheimer’s diseases, chronic kidney disease, and other age-related conditions. When applied to individual cells, the clocks yielded older transcriptomic ages in more than 90 percent of the samples, suggesting that aging is deeply rooted at the cellular level.

In humans, the clocks accurately predicted the lifespans of participants enrolled in a large heart health study. They were also sensitive to environmental factors that affect aging, ticking forward after exposure to radiation or chronic diseases and rewinding after treatments such as young-blood transfusion, a strategy shown to rejuvenate elderly rodents.

An analysis of the genes driving the clocks highlighted many of the usual molecular suspects. Aging turned on genes involved in inflammation, cellular energy disfunction, and senescence—where failing cells leak toxic molecules. Many of these signatures appeared across organs and species, suggesting that core aspects of aging have been conserved in mammals.

These findings are especially valuable for longevity researchers, who often work with rodent models. Despite living a fraction of a human lifespan, aging rodents undergo transcriptomic shifts similar to those found in us. The new clock could easily test their biological age after potential anti-aging treatments, capture the immediate effects, and predict lifespan, long before they die. It could, in theory, speed up aging research and the quest for treatments.

But to be clear, like other aging clocks, it isn’t a crystal ball. Scientists don’t know if the transcriptome changes drive aging or merely reflect its aftermath. The signatures could be capturing overall health and resilience, rather than molecular changes associated with aging per se.

That distinction matters. As we grow older, cells activate a variety of protective genes to counter rising stress, inflammation, and damage. Not every age-related transcriptomic change is harmful. Some changes reflect the body’s attempt to fight back. Because transcriptomes capture only a snapshot in time, scientists still need to differentiate genes that contribute to aging from those that help defend against it and learn how those patterns shift over time.

There’s a broader challenge too. Researchers are building more and more biological clocks using different criteria, and they don’t always agree. One may say you’re far older than another. This highlights “the need for any aging biomarker to be validated carefully,” wrote Magalhães.

The post How Fast Are You Aging? New Genetic Clock May Have the Answer appeared first on SingularityHub.