The Download: seafloor science and military chatbots

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.

Inexpensive seafloor-hopping submersibles could stoke deep-sea science—and mining

Last week, two oblong neon submersibles started to descend nearly 6,000 meters into the Pacific Ocean. Throughout the rest of May, they will map the seafloor in search of critical mineral deposits. 

If all goes well, the vehicles, built by Orpheus Ocean, could help scientists probe the vastly understudied deep sea—and the resources it holds—at a fraction of the cost of existing systems.

But the same submersibles are also attracting deep-sea mining companies, raising concerns about environmental impacts. Find out why they’re drawing so much attention.

—Hannah Richter

The new war room: 10 Things That Matter in AI Right Now 

A new kind of system has entered the war room: conversational AI tools that commanders turn to not just for analysis, but for advice. 

One US defense official told MIT Technology Review that personnel might give these advice engines a list of potential targets to help decide which to strike first. China is commissioning similar tools too.

But as the systems gain traction, they’re also sparking concerns about AI-generated errors, a lack of transparency, and Big Tech gaining undue influence over what information gets seen. 

Here’s how these AI advice engines could impact the battlefield.

—James O’Donnell

The new war room is one of the 10 Things That Matter in AI Right Now, our list of the big ideas, trends, and advances in the field that are driving progress today—and will shape what’s possible tomorrow.

MIT Technology Review Narrated: is fake grass a bad idea? The AstroTurf wars are far from over. 

In 2001, Americans installed just over 7 million square meters of synthetic turf. By 2024, that number was 79 million square meters—enough to carpet all of Manhattan and then some. The increase worries folks who study microplastics and environmental pollution.  

While the plastic-making industry insists that synthetic fields are safe if properly installed, lots of researchers think that isn’t so. 

—Douglas Main 

This is our latest story to be turned into an MIT Technology Review Narrated podcast, which we publish each week on Spotify and Apple Podcasts. Just navigate to MIT Technology Review Narrated on either platform, and follow us to get all our new content as it’s released.

The must-reads

I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.

1 Elon Musk pushed OpenAI to go commercial, its president has testified
Greg Brockman said Musk tried to turn it into a for-profit company years ago. (NYT $)
+ Musk allegedly wanted full control so he could raise $80 billion to colonize Mars. (Reuters $)
+ The Tesla CEO claims he intended for OpenAI to remain a non-profit. (BBC)
+ Here’s what happened in week one of Musk v. Altman. (MIT Technology Review)

2 Google and Meta are building AI agents to rival OpenClaw
Google’s Gemini agent will take actions on the users’ behalf. (Business Insider)
+ Meta’s will be powered by its Muse Spark AI model. (FT $)
+ Hustlers are cashing in on China’s OpenClaw AI craze. (MIT Technology Review)

3 Anthropic will spend $200 billion on Google’s cloud and chips
The investment will be spread across five years. (The Information $)
+ It’s part of a broader AI compute war. (Axios) 

4 DeepSeek is nearing a $45 billion valuation
A state-backed “Big Fund” will lead a new investment round in the company. (FT $)
+ Beijing is pushing to build alternatives to Nvidia and OpenAI. (Bloomberg $)
+ Here’s why DeepSeek’s new model matters. (MIT Technology Review)

5 Anthropic is launching AI agents for banks and financial firms
The 10 tools cover a broad mix of financial services tasks. (WSJ $)
+ They’re part of a push to win over Wall Street. (Bloomberg $)

6 Apple will pay $250 million to settle an AI lawsuit
It was accused of misleading iPhone buyers about Apple Intelligence. (BBC)
+ Some iPhone owners are eligible to receive up to $95. (NYT $)

7 Cheap laptops and phones may be disappearing because of AI demand
 Competition for memory chips is driving up gadget prices worldwide. (The Guardian)

8 Google DeepMind workers in the UK have voted to unionize
As a result of Google’s work with the Pentagon. (Wired $)

9 Pennsylvania is suing Character.AI over chatbots posing as doctors
Investigators say the bots claimed to hold medical licenses. (NPR)
+ How well do AI health tools work? (MIT Technology Review)

10 Scientists created a “living” plastic that destroys itself on command
It could help to eliminate microplastics. (Gizmodo)

Quote of the day

“I want AI to benefit humanity, not to facilitate a genocide.” 

—An anonymous Google DeepMind worker tells the Guardian that Google’s work with the Israel Defense Forces had motivated their vote to unionize.

One More Thing

a tiger shark seen underwater with a camera on its flank

COURTESY OF BENEATH THE WAVES


How tracking animal movement may save the planet

For decades, wildlife researchers have dreamed of building an “Internet of Animals”—a big-data system that monitors and analyzes animal behavior to help us understand the planet. Advances in sensors, AI, and satellite technology are now bringing that vision to reality.

Scientists want the system to track 100,000 sensor-tagged animals. They believe it could reveal how species respond to climate change and ecosystem loss—and even predict environmental disasters. Read the full story on how their idea could save our planet.

—Matthew Ponsford

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.)

+ Master the art of fried chicken with this definitive chef’s guide.
+ Find out why some birds hop and others walk in this breakdown of avian lifestyles.
+ This vintage Hollywood map shows how California’s landscape stood in for everything from the Nile to the Alps.
+ Here’s a fascinating look at the “Flatbed” airplane that was surprisingly efficient on paper but never left the hangar.

Psilocybin-Induced Brain Changes May Explain Therapeutic Effects

Researchers at University of California, San Francisco and Imperial College London have shown that a single dose of psilocybin, the psychedelic compound found in magic mushrooms, causes likely anatomical brain changes that last for up to a month after the experience.

The study, involving healthy volunteers who had never taken a psychedelic, links temporary shifts in brain “entropy”—which is the diversity of neural activity occurring in the brain—to insight. This suggests the psychedelic trip itself is important to the drug’s longer term therapeutic effects.

The researchers found that a high dose of psilocybin led to increased entropy in the minutes and hours after taking the drug. The degree of entropy predicted how much insight, or emotional self-awareness, the participants felt the next day; and this, in turn, forecasted improvements in their sense of wellbeing a month later.

The findings may help to explain psilocybin’s therapeutic effects on conditions such as depression, anxiety, and addiction. “Psychedelic means ‘psyche-revealing,’ or making the psyche visible,” said senior author Robin Carhart-Harris, PhD, the Ralph Metzner distinguished professor of neurology at UCSF. “Our data shows that such experiences of psychological insight relate to an entropic quality of brain activity and how both are involved in causing subsequent improvements in mental health. It suggests that the trip—and its correlates in the brain—is a key component of how psychedelic therapy works.”  Carhart-Harris is senior and corresponding author of the team’s published paper in Nature Communications, titled “Human brain changes after first psilocybin use.”

“Psychedelics have robust effects on acute brain function and long-term behavior but whether they also cause enduring functional and anatomical brain changes is largely unknown,” the authors wrote. Psilocybin is the precursor of the compound psilocin, a serotonin receptor agonist. “Converging evidence supports a role for serotonin 2A receptor  (5-HT2AR) agonism in eliciting the characteristic brain and subjective effects of this and related psychedelics in humans,” the team continued.

For their newly reported study, Carhart-Harris and colleagues carried out an exploratory, placebo-controlled, within-patient study in 28 psychedelic-naïve participants who each received a single, high-dose (25 mg) of psilocybin. The researchers used an assortment of brain imaging and brain measurement techniques, some of which were carried out during the peak of the psychedelic experience, as well as before and one-month after drug administration. “This was an exploratory, hypothesis-generating mechanistic study in healthy volunteers,” the authors noted. None of the 28 people in the study had a diagnosed mental health condition, which gave the scientists greater freedom to do more testing.

In the first part of the experiment the subjects were given a 1 mg dose of psilocybin, which the researchers regarded as a placebo, and were then monitored with EEG, which records brain activity from electrodes on the scalp.  Over the next few weeks, the researchers measured their subjects’ psychological insight, wellbeing, and cognitive ability. They examined brain activity with functional MRI (fMRI) and brain connectivity with diffusion tensor imaging (DTI).

One month after the placebo, the subjects were given 25 mg of psilocybin, a dose capable of eliciting a strong psychedelic trip. During the experience, researchers again measured the subjects’ brain activity with EEG, and in the following weeks they repeated the same tests they had given after the 1 mg dose.

This enabled the scientists to compare the effects of the psychedelic trip on the brain and mind to the effects of the placebo. “The multimodal neuroimaging design allowed us to observe changes in brain function and (potential) anatomy from 1-h (EEG) to 1-month (DTI) after high-dose psilocybin,” they explained.

The investigators found that within 60 minutes of taking the 25 mg dose of psilocybin, EEG revealed higher entropy, suggesting that the brain was processing a richer body of information under the psychedelic. A month later, the researchers looked at their subjects’ brains using DTI, which measures the diffusion of water along neural tracts in the brain, and found that they were denser and had more integrity. This is the opposite of what happens in aging, which makes these tracts more diffuse.

The researchers cautioned that more work needs to be done to better understand the meaning of this finding, but the result is a never-before-seen sign of how psychedelics can change the brain. ”The inclusion of DTI enabled us to test for long-term changes in the integrity of white matter tracts post psilocybin,” the authors stated. “Results revealed decreased axial diffusivity in prefrontal-subcortical tracts 1-month post 25mg psilocybin.”

The day after the 25 mg dose, all but one of the 28 subjects rated the trip as the “single most” unusual state of consciousness they had ever experienced. The remaining person rated it as among their top five. The study participants said they had experienced more psychological insight after taking the 25 mg of psilocybin than they had after the 1 mg placebo.  The subjects also reported increased wellbeing two and four weeks after the study. This was measured from responses to statements such as, “I’ve been feeling optimistic about the future,” and “I’ve been dealing with problems well.”

As the scientists noted in their paper, “A predictive relationship was also found between brain entropy and longer-term mental-health changes—namely, improved wellbeing. Improved wellbeing could be predicted directly from acute increases in brain entropy as early as 1-h post dosing.”

A month after the study the study individuals also scored better on a test of cognitive flexibility.  “Psilocybin seems to loosen up stereotyped patterns of brain activity and give people the ability to revise entrenched patterns of thought,” said first author Taylor Lyons, PhD, a research associate at Imperial College London. “The fact that these changes track with insight and improved well‑being is especially exciting.”

The scientists found that the subjects who had experienced the largest increases in brain entropy in the minutes to hours after taking psilocybin were the most likely to have increased insight the next day and increased wellbeing a month later. The researchers concluded that improved wellbeing was driven by the experience of insight.

The authors suggest that the study findings could improve treatment for people with mental illness using psilocybin, for example, by ensuring that the right dosage is used to produce the right amount of brain entropy to promote insight. “We already knew psilocybin could be helpful for treating mental illness,” Carhart-Harris said. “But now we have a much better understanding of how.”

In their paper the team concluded, “The present multi-modal neuroimaging study in healthy participants sheds light on the brain effects of first-time high-dose psychedelic use and the therapeutic action of psilocybin-therapy, suggesting that therapeutically relevant changes—i.e., improved wellbeing—can be forecast via an acute human brain action, i.e., an entropic brain effect, that is well-known to relate to the psychedelic experience … Results support a role for psychological insight in mediating the causal association between the entropic brain effect and potentially enduring improvements in wellbeing.”

The post Psilocybin-Induced Brain Changes May Explain Therapeutic Effects appeared first on GEN – Genetic Engineering and Biotechnology News.

Muscle Quality and Fat Distribution Predict Mortality Risk Better than BMI

Researchers at the University Medical Center Freiburg in Germany, say that detailed measures of body composition derived from whole-body MRI scans can predict diabetes, cardiovascular events, and mortality risk better than current methods that rely on body mass index (BMI). Using MRI imaging data from more than 66,000 people, the team has developed age-, sex-, and height-adjusted reference standards that show how fat and muscle are distributed across the body and how these patterns relate to health outcomes. Their findings, published in the journal Radiology, show analysis of both the quantity and quality of skeletal muscle, along with where fat is distributed in the body, can provide a more accurate way to determine risk as opposed to weight-based methods alone.

“Many risk scores and treatment decisions still rely on BMI or waist circumference because they are simple to obtain,” said senior author Jakob Weiss, MD, PhD, an interventional radiologist at University Medical Center Freiburg. “But BMI does not reliably reflect a person’s actual body composition.” This is one of the central findings of the study: that individuals with similar BMI values can have markedly different distributions of fat and muscle, which carry different levels of risk for cardiometabolic disease and mortality.

The team’s retrospective study analyzed whole-body MRI scans from 66,608 people using data from the UK Biobank and the German National Cohort collected between April 2014 and May 2022. The cohort had a mean age of 57.7 years and an average BMI of 26.2. Using a fully automated deep learning framework, the researchers quantified multiple body composition measures, including subcutaneous adipose tissue, visceral adipose tissue, skeletal muscle, skeletal muscle fat fraction, and intramuscular adipose tissue. These measures were normalized for age, sex, and height. A score is developed from these data to show how far individuals deviated from a population-adjusted reference.

“Whole-body MRI–derived BC (body composition) z-scores were used to identify at-risk individuals and predict cardiometabolic outcomes and mortality beyond traditional risk factors.” They then used the z-score categories to assess associations and clinical outcomes.

Their data showed that individuals with high visceral fat had a 2.26-fold increased risk of developing diabetes. High intramuscular fat was associated with a 1.54-fold increased risk of major adverse cardiovascular events, while low skeletal muscle was linked to a 1.44-fold increase in all-cause mortality.

The deep learning system used to develop the risk profiles was trained and evaluated against radiologist-defined reference standards, allowing it to extract volumetric measurements across the entire body rather than relying on single cross-sectional slices. This method allowed the researchers to capture meaningful variations in muscle quality and fat distribution that are not visible through other techniques.

“Manual BC measurement in large-scale imaging datasets is prohibitively time-consuming,” the researchers wrote. “However, recent advances in deep learning have enabled fully automated, accurate, and efficient quantification from cross-sectional imaging.” This capability allowed the team to construct reference curves reflective of how body composition changes with age and differs between men and women.

Importantly, the research shines a light on the limitations of using BMI to determine future risk. Because BMI is calculated using only two metrics, height and weight, it does not distinguish between fat and muscle or account for where fat is stored. Because of this, two people with the same BMI may have very different levels of visceral fat or muscle mass, factors that can lead to different to different health risks. The researchers showed that deviations in these specific components, captured via their MRI-based z-scores, were predictive of outcomes even after accounting for traditional risk factors.

“It’s not only how much muscle you have, but also it’s the quality of that muscle,” said first author Matthias Jung, MD, a radiologist at University Medical Center Freiburg. “Knowing the volume of intramuscular fat gives us a window into muscle quality that other methods like BMI, bioelectrical impedance analysis, or DEXA can’t easily provide.” This distinction is relevant because intramuscular fat is linked to metabolic dysfunction and cardiovascular risk.

The study also produced a web-based calculator that allows clinicians and researchers to compare individual patient data with population-based reference values. According to Weiss, this tool could be applied to existing imaging studies. “A dedicated whole-body MRI is not necessarily required. If a routine CT or MRI body scan already exists, the information can be extracted for benchmarking against the reference values,” he said.

The study has limitations, including a cohort of primarily White Western European adults, which may impact the generalizability of the findings. The researchers also pointed out that whole-body MRI is not routinely performed in clinical practice, although they provided reference values for commonly imaged regions such as the chest, abdomen, and pelvis to address this.

The team will continue their work by seeking to validate the reference curves in clinical populations and exploring their use in predicting treatment outcomes, including toxicity, survival, and recurrence in cancer patients. The team also plans to develop disease-specific reference values for broader patient groups to broaden the use of body composition analysis into clinical care.

The post Muscle Quality and Fat Distribution Predict Mortality Risk Better than BMI appeared first on Inside Precision Medicine.

AI in Endoscopic Transsphenoidal Surgery

Conditions: Pituitary Adenoma

Interventions: Device: Live intra-op AI analysis of endoscopic video feed, with output displayed on supplementary monitor

Sponsors: University College, London; University College London Hospitals

Not yet recruiting

TRACS Enables Strain-Level Tracking of Microbial Transmission

Tracking microbes is challenging, particularly when there are coexisting strains of the same species within metagenomic data. However, overcoming that challenge is important for inferring transmission of both pathogenic and commensal microbes.

A new tool, called TRAnsmision Clustering of Strains (TRACS), distinguishes between closely related bacterial strains. The “highly accurate algorithm” can be used for “estimating genetic distances between strains at the level of individual single nucleotide polymorphisms, which is robust to intra-species diversity within the host.”

Researchers used the TRACS tool to map the transmission of SARS-CoV-2, Streptococcus pneumoniae, and Plasmodium falciparum (the causative agent of malaria) across different populations. The tool may play an important role in infection prevention, outbreak response, and the development of treatments designed to help the human microbiome fight infection. They note that this tool can be used across microbial kingdoms to uncover strain dynamics.

“Traditionally, this has been very difficult for us to achieve, yet it is incredibly important to know, as people can carry several slightly different versions or strains of the same species at once, which makes it challenging to understand how microbes move between individuals,” notes Gerry Tonkin-Hill, PhD, group leader at the the Peter MacCallum Cancer Centre and the Peter Doherty Institute at the University of Melbourne, Australia. “Using this new technology, we can now overcome this challenge and gain a clearer picture of how microbes are shared between people. This will give us a better understanding of how microbes spread to help us prevent infection in vulnerable populations, like our cancer patients.”

This work is published in Nature Microbiology in the paper, “Strain-level transmission inference across multi-kingdom metagenomic data using TRACS.”

Being able to track the spread of pathogens using genomics has become a major tool in public health and can help inform new ways to prevent transmission. Additionally, it can help understand more about how lifestyle and environmental factors are involved in the transmission of these pathogens, and their role in the microbiome.

Currently, genomic tools used to track multiple bacterial species do not have the speed and flexibility required for routine public health monitoring and can struggle to distinguish between samples transmitted recently and those transmitted years ago. Furthermore, it can be difficult to continuously add in new samples, making real-time surveillance difficult.

The TRACS algorithm identifies and analyzes Single Nucleotide Polymorphisms (SNPs) to estimate how closely related the pathogens are, and if they are likely to have recently been transmitted. This approach allows for the continuous integration of new samples, making it an ideal tool for accurately identifying transmission networks and ruling out transmission events in ongoing public health applications.

In this new study, the team used TRACS to map pathogen transmission networks across three different populations, all of which had different genomic data. They applied it to SARS-CoV-2 data from U.K. hospitals, deep population sequencing data of Streptococcus pneumoniae and single-cell genome sequencing data from malaria patients infected with Plasmodium falciparum. They found that the tool was able to identify different pathogens in one sample and infer where these were each transmitted.

They also used TRACS to study how microbes are passed from mothers to infants and found that one beneficial bacterium, Bifidobacterium breve, persisted in infants longer than previously recognized, something that previous methods have missed.

More superficially, the authors note that “applying TRACS to gut metagenomic samples from a mother–infant cohort revealed species-specific transmission rates and identified increased the persistence of Bifidobacterium breve in infants, a finding previously missed owing to the presence of multiple strains.”

“This research could support the development of new treatments that use beneficial microbes to improve health,” notes Trevor Lawley, PhD, group leader at the Wellcome Sanger Institute. “By understanding exactly how microbes move between people and which of them are more likely to thrive in their microbiome, we could design better ways to increase helpful gut microbes and investigate whether there are ways to use these to help prevent infections, opening the door to safer healthcare environments and new microbiome-based therapies.”

The post TRACS Enables Strain-Level Tracking of Microbial Transmission appeared first on GEN – Genetic Engineering and Biotechnology News.

Optical Pooled CRISPR Screen Reveals Regulators of NF-κB Dynamics in Human Cells



Image of Tilmann Buerckstuemmer, PhD

Tilmann Buerckstuemmer, PhD

CSO
Myllia Biotechnology

Panelist

Image of Tilmann Buerckstuemmer, PhD

Tilmann Buerckstuemmer, PhD

Tilmann Buerckstuemmer, PhD, is a CRISPR enthusiast since the early days of CRISPR. Originally trained as a biochemist, he joined Haplogen as principal scientist and later became their CSO. Following the acquisition by Horizon Discovery, Tilmann served as director of research and development and later as head of innovation, where he oversaw the company’s technology platform and innovation agenda. In 2018, he co-founded Myllia Biotechnology which focuses on single-cell CRISPR screens. He is also the CEO of bit.bio discovery, a joined venture between Vienna-based Myllia Biotechnology and Cambridge-based bit.bio. Tilmann is passionate about science and enjoys working with multi-disciplinary and multi-national teams.



Image of Jens Durruthy Durruthy, PhD

Jens Durruthy Durruthy, PhD

Director of Product Management
Element Biosciences

Panelist

Image of Jens Durruthy Durruthy, PhD

Jens Durruthy Durruthy, PhD

Jens Durruthy Durruthy, PhD, is the director of product management at Element Biosciences. Prior experience includes a decade at 10x Genomics, where he developed and oversaw the product portfolio for Chromium products. Jens held the position of LSA Bio/Genomics Fellow at Life Science Angels, conducting extensive research on investment opportunities in biotech and genomics startups, and has worked in various consulting roles, focusing on product development and market analysis. Educational credentials include a PhD in biomedical engineering from Stanford University and a diploma in medical biotechnology from Technische Universität Berlin.



Broadcast Date: 

  • Time: 

Integrated pooled CRISPR screening linked to imaging readouts accelerate target identification and functional characterization of signaling pathways. A good example of this can be found in studies of NF-κB signaling, which is central to inflammatory responses and driven by rapid nuclear translocation of the p50/p65 complex to activate transcriptional programs following cytokine stimulation.

In this GEN webinar, Tilmann Buerckstuemmer, PhD, CSO at Myllia Biotechnology will show how high-throughput pooled CRISPR screening combined with cell painting readouts characterized important signaling pathways using NF-κB nuclear translocation as a case study. During the webinar, you will learn how the AVITI24™ platform from Element Biosciences profiled ~440,000 cells in a pooled CRISPR screen targeting 195 genes. Linking genetic perturbations to p65 subcellular localization and cell painting features in a single workflow enabled identification of known pathway components, uncovered regulatory roles for chromatin-modifying complexes, and improved interpretation of phenotypic outcomes using morphological features.

Key takeaways include:

  • Strategies for linking CRISPR perturbations to protein localization and morphological features at single-cell resolution
  • Identification of hitherto poorly characterized chromatin modifying complexes in regulating NF-κB signaling
  • The value of multimodal readouts, including morphology, in adding depth and confidence to recovered biology
  • How this approach supports mechanism-of-action studies and enables identification of both positive and negative regulators of signaling pathways

A live Q&A session will follow the presentation offering you a chance to pose questions to our expert panelists.

Produced with support from:

Element Bio logo

The post Optical Pooled CRISPR Screen Reveals Regulators of NF-κB Dynamics in Human Cells appeared first on GEN – Genetic Engineering and Biotechnology News.

Late-life difficult-to-treat depression and dementia subtypes: a naturalistic cohort study using electronic health records

IntroductionLate-life difficult-to-treat depression (LL-DTD) and dementia frequently coexist in later life, but it remains unclear whether clinical and sociodemographic characteristics, as well as medication exposure patterns, differ across dementia subtypes among older adults with both conditions.MethodsWe analysed anonymised electronic health records from a south London catchment area. We included patients aged ≥60 years at first recorded depression diagnosis with a dementia diagnosis. LL-DTD was defined as inadequate response to ≥2 antidepressant trials. Dementia diagnoses were classified as Alzheimer’s disease (AD), vascular dementia (VD), mixed AD/VD, dementia with Lewy bodies (DLB), or other/unspecified dementia. Around the index depressive episode, we captured sociodemographics, depressive symptoms, physical comorbidity, and medication indicators. We used multivariate logistic regression to examine cross-sectional correlates distinguishing dementia subtypes (with AD as reference) within the LL-DTD and dementia sample. We conducted stratified analyses comparing non-AD versus AD dementia by the temporal order of dementia and depression diagnoses.ResultsAmong 890 older adults with LL-DTD and dementia, AD was the most common subtype (33.9%), followed by mixed AD/VD (22.5%), VD (22.2%), other/unspecified dementia (15.2%), and DLB (6.2%). Depressive symptom profiles and psychotropic treatment history were broadly similar across subtypes. Compared with AD, VD was associated with greater functional impairment, while greater physical comorbidity burden was more evident in VD and mixed AD/VD.DiscussionSomatic multimorbidity and functional impairment provided the clearest clinical separation between subgroups, while depressive symptom patterns and medication exposure appeared largely non-specific across dementia subtypes. This underscores the importance of multimorbidity and physical health burden in understanding heterogeneity of dementia outcomes in LL-DTD.