<![CDATA[New data shows stable symptoms when switching from atypical antipsychotics to Cobenfy using cross-titration.]]>
<![CDATA[In the wake of the FDA fast-track of psychedelic therapies, we want to hear your clinical pearls for our May theme.]]>
The Download: a new Christian phone network, and debugging LLMs
This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology.
A new US phone network for Christians aims to block porn and gender-related content
A new US-wide cell phone network marketed to Christians is set to launch next week. It blocks porn using network-level controls that can’t be turned off—even by adult account owners.
It’s also rolling out a filter on sexual content aimed at blocking material related to gender and trans issues, optional but turned on by default across all plans.
The trouble is, many websites don’t fit neatly into one category. That leaves its maverick founder with broad, subjective control over what is allowed or banned. Read the full story.
—James O’Donnell
This startup’s new mechanistic interpretability tool lets you debug LLMs
The San Francisco–based startup Goodfire has released a new tool, Silico, that lets researchers peer inside an AI model and adjust its parameters during training. It could give users more control over how this technology is built than was once thought possible.
The goal is to make building AI models less like alchemy and more like a science. Using a technique called mechanistic interpretability, Silico maps the neurons and pathways inside a model and lets developers tweak them to reduce unwanted behaviors or steer outputs.
By exposing the “knobs and dials,” Goodfire hopes to bring AI training closer to traditional software engineering. Read the full story.
—Will Douglas Heaven
With mass firing, Trump deals a fresh blow to American science
This past week delivered another gut punch for science in the US. This time, the target was the National Science Foundation—a federal agency that funds major research projects to the tune of around $9 billion. On Friday, the 22 scientists overseeing those efforts were all fired.
Since 2025, the NSF has faced budget cuts, grant terminations, and mass firings, with staff numbers down sharply and many ambitious projects grinding to a halt. The result is a major shift in how American science is funded and governed. Discover what it means, and what’s next.
—Jessica Hamzelou
This article first appeared in The Checkup, MIT Technology Review’s weekly biotech newsletter. To receive it in your inbox every Thursday, and read articles like this first, sign up here.
China’s open-source bet: 10 Things That Matter in AI Right Now
Silicon Valley AI companies follow a familiar playbook: keep the models behind an API and charge for access. China’s leading AI labs are playing a different game, releasing “open-weight” models that developers can download, adapt, and run on their own hardware.
That approach went mainstream after DeepSeek open-sourced its R1 model, which matched top US systems at a fraction of the cost. It also won something subtler: goodwill with developers. A growing cohort of Chinese labs is now following the same blueprint.
As AI shifts from hype to deployment, open-source models are making the future of AI more multipolar than Silicon Valley expected. Read the full story.
—Caiwei Chen
China’s open-source bet is one of the 10 Things That Matter in AI Right Now, our list of the biggest ideas, trends, and advances in AI today. We’re unpacking one item from the list each day here in The Download, so stay tuned.
The must-reads
I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.
1 Elon Musk has admitted that xAI trained Grok on OpenAI models
“Distillation” is standard practice in AI, despite being legally dubious. (Wired $)
+ The White House has accused Chinese firms using distillation of theft. (BBC)
+ American labs are widely assumed to use similar techniques. (TechCrunch)
2 A “de-extinction” startup wants to resurrect a long-lost antelope
Colossal Biosciences wants to bring back the bluebuck. (Axios)
+ The company is using genomic editing to revive the animal. (Gizmodo)
+ It previously claimed to have cloned red wolves. (MIT Technology Review)
3 An OpenAI model outperformed ER doctors at diagnosing patients
By analyzing health records data and information provided to physicians. (NPR)
+ But it still must be proven in real-world clinical trials. (Vox)
4 Scientists are trying to power AI data centers with tiny nuclear reactors
They could provide a new way to meet AI’s energy demands. (Gizmodo)
+ We did the math on AI’s energy footprint. (MIT Technology Review)
5 Spotify has started verifying human artists
A new badge will distinguish them from AI. (The Guardian)
+ Spotify has faced criticism for its handling of AI. (BBC)
6 The US is backing a Congolese railway to break China’s grip on critical minerals
The old railroad is key to the race for critical metals in Africa. (Rest of World)
+ The US is also searching for alternative sources. (MIT Technology Review)
7 Huawei is set to overtake Nvidia in China’s AI chip market
It’s expected to capture the largest market share this year. (FT $)
8 Japan is building cardboard drones for the battlefield
The flatpack designs are cheap, disposable, and built at scale. (404 Media)
9 The more young people use AI, the more they hate it
Research shows that Gen Z doesn’t trust GenAI. (The Verge)
10 A new organoid can menstruate—and show how tissue repairs itself
It’s revealing how the uterus can shed without scarring. (Nature)
Quote of the day
“I suspect that there are a number of people who do not want to put the future of humanity in Mr Musk’s hands. But we’re not going to get into that.”
—Judge Gonzalez Rogers rebukes attempts by Elon Musk’s lawyer to focus on AI’s existential risks as part of his lawsuit against OpenAI, the New York Times reports.
One More Thing
TMY350 VIA WIKIMEDIA COMMONS
This rare earth metal shows us the future of our planet’s resources
The materials we need to power our world are shifting from fossil fuels to energy sources that don’t produce greenhouse gas emissions.
Take neodymium, a rare earth metal used in powerful magnets that power everything from smartphones to wind turbines. Its story reveals many of the challenges we’ll likely face across the supply chain in the coming century and beyond.
The question isn’t whether we’ll run out, but how we extract, process, use, and recycle these materials as technology keeps changing. Find out what it reveals about the future of our planet’s resources.
—Casey Crownhart
We can still have nice things
A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line.)
+ Here’s a fascinating visual history of exploring the dark side of the Moon.
+ This interactive map lets you compare the actual dimensions of our world.
+ These five tiny homes are proof you don’t need a massive footprint to live with style.
+ Explore the history of “Control Room Green” and why it was the default choice for the Cold War’s highest-stakes environments.
Top image credit: Stephanie Arnett/MIT Technology Review | Adobe Stock
Examining circadian rhythm dysregulation using actigraphy among treatment-seeking individuals with alcohol use disorder
BackgroundIdentifying factors predictive of relapse in patients with alcohol use disorder (AUD) following a period of abstinence and/or treatment is essential to discover effective treatment plans for this disease. Previous evidence found that individuals with AUD who relapsed had lower sleep regularity scores than those who did not relapse. This analysis aimed to extend previous work to explore the relationship between circadian rhythms and relapse.MethodsTreatment-seeking individuals with AUD (n = 126) were admitted to an inpatient treatment program for approximately 28 days and, upon discharge, wore Philips Respironics Actiwatches® for the subsequent 4 weeks during an unprotected environment. A subset of these participants wore the devices prior to discharge for up to 7 days inpatient (n = 36). Relapse status was assigned if a participant consumed any alcohol during the outpatient period of data collection. Inpatient and outpatient circadian rhythm nonparametric statistics were calculated for all participants including weekly interdaily stability (IS) and intradaily variability (IV), and daily most active 10 h (M10), least active 5 h of the day (L5), relative amplitude (RA), and the wake time. Linear and logistic generalized mixed models were fitted to estimate the effect of discharge on circadian rhythms, the effect of preceding circadian rhythms on the probability of relapse, and the effect of relapse on circadian rhythms. All analyses accounted for within-patient repeated measurements.ResultsThe final cohort size was n = 103 for the outpatient subset and n = 36 for the inpatient subset after actigraphy data filtering. Participants were 48.6 ± 11.3 years of age, and 32% were female. A total of 26 (25.2%) participants relapsed. There were significant decreases in IV and L5 and increases in M10, RA, and wake time between inpatient and outpatient settings, whereas IS did not substantially differ between the two settings. Following relapse, there was a moderate decrease in hourly IV and an increase in (later) wake time of ~1 h; no other circadian variables were significantly predictive of relapse.ConclusionOverall, circadian rhythms shifted after discharge but were not predictive of relapse. Instead, relapse was followed by a delayed average wake time and a moderate reduction of daily activity patterns. These results highlight the potential value of monitoring circadian changes as indicators of relapse occurrence rather than relapse risk.
A game-theoretic framework for multimodal information utilization under heterogeneous processing environments in neuroscience and perception science
Multimodal data integration is increasingly central to neuroscience and perception science, where heterogeneous signals such as behavioral responses, sensory inputs, electrophysiological recordings, neuroimaging measurements, and computational representations must be jointly interpreted. Based on the realistic background, there is a core theoretical problem that needs further research: under what heterogeneous processing conditions does enhanced multimodal information utilization produce meaningful gains, when does it become strategically necessary, and when does it generate only limited benefits relative to its cost? To clarify this core problem, this study develops a conceptual game-theoretic framework in which information utilization is treated not as a universally beneficial technical upgrade, but as a conditional strategic choice shaped by signal heterogeneity, information asymmetry, integration cost, and differential decision influence across actors. Within this framework, we compare three endogenous strategic profiles—no enhanced information utilization, unilateral information enhancement, and bilateral information enhancement—across multiple heterogeneous environments. The analysis results show that the value of multimodal information utilization is fundamentally environment-dependent. In highly homogeneous environments, additional information processing yields little marginal benefit and is therefore not sustained in equilibrium. In moderately heterogeneous environments, however, multimodal information utilization emerges as a strategically necessary response because it reduces mismatch, improves alignment, and stabilizes decision outcomes. In more asymmetric environments, stronger decision agents capture a disproportionate share of the gains from enhanced information utilization and increasingly rely on differentiated strategic responses, whereas weaker agents adopt more defensive and uniform strategies. In highly dominated environments, the marginal value of additional information utilization declines again because structural dominance itself already secures most attainable advantages. These findings contribute to multimodal neuroscience and perception science by clarifying that the consequences of information utilization depend not only on fusion efficiency, but also on environmental structure, asymmetry, and the distribution of strategic power.
Computational learning phenotypes are not related to individual differences in resting-state fMRI connectivity
People learn from experience, but with considerable individual differences in the degree and type of behavioral adjustments resulting from a given experience. Error driven learning rules provide an elegant framework for explaining both learning behavior and its neural signatures; however, implementing them requires carving the world into so-called “latent states”, that serve as substrates for learning, meaning that the same learning algorithm can produce different sorts of learning given different state representations. Recent theoretical and behavioral work hints that individual differences in learning may reflect differences in how individuals carve their environment into states, with some individuals combining multiple temporal contexts into a single state and others separating these contexts into individuated latent states. Here, we develop a behavioral paradigm and modeling framework to test this idea directly and show in a large cohort of human participants that individuals can be classified into groups according to whether and how they carve temporal contexts into latent states. These behavioral phenotypes impact continual learning, specifically the degree to which individuals avoid interference at context changes or are able to reuse information when encountering a familiar context. We tested whether these behavioral phenotypes related to individual differences in underlying brain connectivity, as measured by resting state-fMRI, but found no evidence for such a relationship. Taken together, this work suggests that learning differences across individuals are attributable to differences in underlying state representations that are not predicted by underlying resting state brain connectivity.
Context-tuned strategies for marker selection precision in neuronal studies
Marker selection precision in neuronal studies is critical for reliable neuron identification. However, it largely depends on the experimental context. Variations in neuronal marker specificity across experimental models, neuronal maturation stages, and neurotransmitter phenotypes have highlighted the vitality of implementing “context-tuned” strategies in marker selection. Neuronal markers arise from canonical protein-coding genes, non-coding RNAs (ncRNAs), including microRNAs (miRNAs), isoform-specific variants, neurotransmitters, and numerous metabolic signatures. Here, we emphasize protein-coding genes as markers because of their wide availability, ease of interpretation, and compatibility with standard detection methods like qPCR, in situ hybridization, immunocytochemistry, and Western blotting. They are also directly linked to cellular structures, signaling pathways, functional importance, and are adaptable across different platforms. We aim to guide the strategic selection and application of neuronal markers to maximize accuracy and interpretive confidence across diverse experimental systems. The review addresses the molecular origin and nature of neuronal markers, their specific applications, including distinguishing neuronal from non-neuronal cells in tissue or histological preparations, identifying neurotransmitter phenotypes in neuronal cultures and tissues, evaluating neuronal maturity in progenitor-derived systems, discriminating between immature and fully differentiated neurons in vitro, and detecting neurons alongside other neuronal or non-neuronal subtypes in mixed populations. Furthermore, it emphasizes positive and negative marker strategies, accounting for developmental timing, cellular specificity, model system differences, and rigorous exclusion of unintended cell types. Through this comprehensive review, we deliver a simplified reference for neuroscientists seeking to enhance the accuracy, specificity, and reproducibility of their neurobiological studies.
Timing of exercise differentially modulates fear memory and hippocampal neurotransmitters in male rats
Exercise promotes neurogenesis and enhances memory consolidation while reducing the retention of aversive memories and anxiety-like behaviors. While our previous work found that acute exercise alters neurotransmitter concentrations, including dopamine and serotonin, in a time-of-day-dependent manner, the long-term effects of chronically timed exercise on neurotransmitter dynamics and behavioral phenotypes remain unclear. To examine whether the daily timing of a chronic exercise intervention modulates its impact on neurotransmitter profiles and fear responses, male rats were conditioned using a Pavlovian contextual fear approach, then assigned to a 4-week treadmill exercise intervention performed during the early (ZT14) or late (ZT22) active phase or a time-matched sham-exercise control group. One day after completing training, rats underwent a context retrieval test in the middle of active phase (ZT18), and hippocampal neurotransmitters were quantified using UPLC–MRM/MS. Rats subjected to sham-exercise at ZT22 exhibited higher freezing than sham-exercised rats at ZT14, whereas exercise interventions at ZT22 selectively attenuated freezing. Histamine, acetylcholine, and GABA exhibited significant exercise × time interactions. Direct neurotransmitter–freezing correlations were weak after false discovery rate control, consistent with a network-level reorganization rather than a single transmitter driver. These findings suggest that vulnerability to aversive memory expression can be buffered by exercise, if timed appropriately, and that exercise reshapes hippocampal neuromodulatory tone in a circadian–phase–dependent manner, supporting the potential of exercise timing as a chronotherapeutic strategy to enhance stress resilience and mental wellbeing.
RecoveryWorks: vocational support services for recovery capital growth
This paper introduces RecoveryWorks, a multi-system recovery and vocational support program designed to expand the therapeutic landscape of recovery by increasing access to career development resources that support meaningful employment. In contrast to vocational approaches that prioritize rapid job placement, RecoveryWorks embeds work within a recovery-oriented system of care. By linking the underutilized human capital of individuals in recovery to community resources, the program aims to generate reciprocal gains in personal and community recovery capital over time.
Meta-analysis of the effects of exercise intervention on physical health in individuals undergoing compulsory isolation
BackgroundPhysical health is the basic indicator to evaluate the health of drug addicts after the process of drug rehabilitation. In order to better improve the deficiency degree of physical health of drug addicts, it is necessary to carry out a systematic review.ObjectiveTo explore the effects of exercise intervention on the physical health of individuals undergoing compulsory drug rehabilitation using Meta-Analysis, aiming to provide evidence-based support for improving their physical health.MethodsRandomized controlled trials (RCTs) published between 2019 and December 2024, examining the impact of exercise intervention on the physical health of compulsory detoxification individuals, were retrieved from databases including Web of Science, PubMed, Cochrane Library, Medline, China National Knowledge Infrastructure (CNKI), Wanfang Data, and VIP Chinese Journal Database. The quality of included studies was assessed using the Cochrane risk-of-bias assessment tool. RevMan 5.4 software was employed for heterogeneity testing, effect size synthesis (using mean difference [MD] and 95% confidence interval [CI]), and generation of forest plots, funnel plots, and quality assessment diagrams. Subgroup analyses were performed to evaluate sensitivity and heterogeneity of the included studies.ResultsExercise intervention effectively improved the physical health of compulsory drug rehabilitation individuals, particularly in physical fitness indicators: sit-and-reach test [MD = 3.92, 95%CI = (3.23, 4.62), P<0.001], single-leg standing with eyes closed [MD = 7.03, 95%CI = (6.05, 8.02), P<0.001], grip strength [MD = 1.23, 95%CI=(0.06, 2.39), P = 0.04], and choice reaction time [MD=-0.03, 95%CI=(-0.05, -0.01), P = 0.002]. Improvements in physical function were also observed; however, the increase in vital capacity [MD = 86.81, 95%CI=(-1.56, 175.17), P = 0.05] did not reach statistical significance.ConclusionThis meta-analysis provides evidence that exercise intervention significantly improves specific physical health deficits—namely flexibility (sit-and-reach), balance (single-leg stance), muscular strength (grip strength), cardiopulmonary function (vital capacity), and sensorimotor coordination (choice reaction time)—in individuals undergoing compulsory rehabilitation. It is recommended to adopt a combination of aerobic and traditional fitness exercises, with at least 3 sessions per week, each lasting no less than 40 minutes, and a duration of over 12 weeks, providing scientific evidence for drug rehabilitation practices. These indicators were selected because they directly reflect the multisystem damage (muscular, neural, and cardiorespiratory) caused by chronic substance use. However, this study acknowledges the limitation that psychological and neurocognitive outcomes (e.g., cravings, mood, executive function), which are crucial in addiction treatment, were not included in the eligibility criteria and systematic analysis. The follow-up research will combine physical and psychological indicators to conduct a comprehensive evaluation of the intervention effect of exercise on drug rehabilitation.Systematic review registrationhttps://www.crd.york.ac.uk/prospero/, identifier CRD420251029820.