<![CDATA[Study links schizophrenia’s earlier onset to higher genomic deletion CNV burden, showing future potential for personalized care.]]>
Human Shadows in Machine Minds: Quantitative Study Interpreting AI Responses to the Rorschach Test
<strong>Background:</strong> Multimodal large language models (LLMs) can produce humanlike descriptions of images and emotionally colored dialogue, which motivates research on how psychological assessment methods might be adapted to evaluate model behavior under ambiguity. Projective tests such as the Rorschach inkblot test have rarely been applied to LLMs. <strong>Objective:</strong> This study assessed the feasibility of administering a full Rorschach protocol to multimodal LLMs and descriptively compared response features by using established Rorschach coding categories. <strong>Methods:</strong> We presented all 10 standard Rorschach cards to 3 multimodal LLMs (GPT-4o, Grok 3, and Gemini 2.0 Flash Thinking). We used the standard prompt (“What might it be?”) and a prespecified fallback prompt for models that did not provide codable responses. We conducted an inquiry phase and coded responses using the Exner Comprehensive System, summarizing response count (R), location (W and D), determinants (eg, F, M, and C), and human-related content. As an exploratory step, we also prompted an additional LLM (Anthropic 3.7) to summarize and count response features and compared these outputs with manual tallies. For GPT-4o, we additionally tested image generation of its interpretations. <strong>Results:</strong> GPT-4o completed the administration using the standard prompt; Grok 3 and Gemini required the fallback prompt. The total number of responses was 15 for GPT-4o, 10 for Grok 3, and 20 for Gemini. GPT-4o and Grok 3 produced mainly whole-blot responses (13/15, 86.7% and 9/10, 90%, respectively), whereas Gemini produced mainly common-detail responses (16/20, 80%). Human movement determinants were more frequent in GPT-4o (7/15, 46.7%) and Grok 3 (3/10, 30%) than in Gemini (1/20, 5%). Human-themed contents occurred 46.7% (7/15), 50% (5/10), and 20% (4/20) of the time, respectively. Anthropic 3.7 reproduced some counts but showed errors in response and determinant tallies for 2 of the 3 models. <strong>Conclusions:</strong> Multimodal LLMs can generate Rorschach-like narratives that map onto standard coding categories, but outputs are sensitive to prompting and platform constraints and should not be interpreted as evidence of a model “inner world.” LLM-assisted coding showed limitations. The emergent behavior of LLMs was examined using the Rorschach test, and their response phenotype, based on this analysis, showed deviations from typical human normative patterns. Future work should use controlled sampling, repeated administrations, and stimulus sets less likely to have been seen during training.
The Download: Musk and Altman’s legal showdown, and AI’s profit problem
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.
Elon Musk and Sam Altman are going to court over OpenAI’s future
Elon Musk and OpenAI CEO Sam Altman head to trial this week in a case with sweeping consequences. Ahead of OpenAI’s IPO, the court could rule on whether the company can exist as a for-profit enterprise. It could even oust its leadership.
Musk, an OpenAI co-founder, claims he was deceived into bankrolling the firm under false pretenses. He’s seeking $134 billion in damages, the removal of Altman and president Greg Brockman, and the company’s restoration to a non-profit.
Find out how the trial could upend the global AI race.
—Michelle Kim
The missing step between hype and profit
In a celebrated South Park episode, a community of gnomes sneak out at night to steal underpants. Why? The gnomes present their pitch deck. “Phase 1: Collect underpants. Phase 2: ? Phase 3: Profit.” It’s a business plan that captures the current state of AI.
Companies have built the tech (Step 1) and promised transformation (Step 3). But how they get there is still a big question mark. Read about the potential paths forward.
—Will Douglas Heaven
This story originally appeared in The Algorithm, our weekly newsletter giving you the inside track on all things AI. Sign up to receive it in your inbox every Monday.
Welcome to the era of weaponized deepfakes
For years, experts have warned that deepfakes could be deployed in malicious ways. These dangers are now here.
Cheap, accessible models now produce weaponized deepfakes—from sexually explicit images to political propaganda—that look startlingly real. They’re already inciting violence, changing minds, and sowing mistrust, with women and marginalized groups disproportionately affected.
Experts fear that they’re cratering trust and critical thinking. Here’s why they’re alarmed.
—Eileen Guo
Weaponized deepfakes are on our list of the 10 Things That Matter in AI Right Now, MIT Technology Review’s guide to what’s really worth your attention in the busy, buzzy world of AI.
The must-reads
I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.
1 OpenAI has ended its exclusive partnership with Microsoft
The new deal allows OpenAI to court rivals such as Amazon. (Reuters $)
+ Microsoft will still license OpenAI’s tech, but no longer exclusively. (NYT $)
+ OpenAI is missing key growth targets ahead of its IPO. (WSJ $)
2 Google has signed a classified AI deal with the Pentagon
It permits AI use for “any lawful government purpose.” (The Information $)
+ Over 600 Google workers had called for a block on the deal. (QZ)
+ AI firms are set to train military versions of their models on classified data. (MIT Technology Review)
3 The EU has told Google to open Android to AI rivals
It wants to end Gemini’s built-in advantage. (Ars Technica)
+ Google calls the move an “unwarranted intervention.” (WSJ $)
+ A final decision is expected by the end of July. (Reuters $)
4 OpenAI is reportedly developing an AI-first smartphone
It would replace apps with agents. (TechCrunch)
+ Qualcomm and MediaTek may be developing its processors. (Gizmodo)
5 A brain implant for depression is moving into human testing
The FDA has approved a human study of the device. (Wired $)
+ BCIs have thus far struggled to reach the market. (MIT Technology Review)
6 A populist backlash against AI is gaining momentum in rural America
From Indiana to Idaho, voters are pushing back against the technology. (NYT $)
+ Anti-AI protests are expanding worldwide. (MIT Technology Review)
7 DeepSeek has priced its new model 97% below OpenAI’s GPT-5.5
It aims to attract more enterprises, developers, and agent-based users. (SCMP)
+ Here are three reasons why DeepSeek V4 matters. (MIT Technology Review)
8 AI now generates a third of new websites
A study found it’s making the web more cheery and less verbose. (404 Media)
9 Top talent is leaving Big Tech to launch their own AI startups
Meta, Google, and OpenAI are facing a brain drain. (CNBC)
10 Taylor Swift is trademarking her voice and image
The Grammy winner has been the target of numerous deepfakes. (NBC News)
+ A growing number of celebrities are fighting AI with trademarks. (BBC)
Quote of the day
“The reality is people don’t like him.”
—Judge Yvonne Gonzalez Rogers reacts to prospective jurors confessing their negative views of Elon Musk ahead of his legal battle with Sam Altman, The Verge reports.
One More Thing
How covid conspiracy theories led to an alarming resurgence in AIDS denialism
When Joe Rogan falsely declared that “party drugs” were an “important factor in AIDS,” several million people were listening. He also asserted that AZT, the earliest drug used to treat AIDS, killed people “quicker” than the disease itself—another claim that has been disproven.
Such comments illustrate an unmistakable resurgence in AIDS denialism: a false collection of theories arguing either that HIV does not cause AIDS or that there is no such thing as HIV at all. By the dawn of the millennium, these claims had largely fallen out of favour. That changed when the coronavirus arrived.
Follow the digital path from Covid skepticism to the return of a deadly conspiracy theory.
—Anna Merlan
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.)
+ Explore the planets from your laptop with this live sky map.
+ This marathon DJ set from Daphni is an incredible journey through electronic music.
+ NASA’s stunning Artemis II wallpapers bring a high-res piece of deep space to your phone.
+ This fascinating GPS explainer breaks down how your phone figures out exactly where you are.
A structured narrative review on VNS-treated drug-resistant epilepsy: EEG markers, neurochemical mechanisms, and future biomarker-driven computational directions
Vagus nerve stimulation (VNS) is an established adjunctive therapy for drug-resistant epilepsy. Experimental and clinical evidence indicates that its therapeutic effects involve distributed brain networks and multiple neurochemical pathways. Electroencephalography (EEG) has been widely used to characterize VNS-related neurophysiological changes, including alterations in conventional oscillatory activity, functional connectivity, and, more recently, aperiodic spectral components such as the spectral exponent and spectral offset. However, these EEG findings are often interpreted without sufficient consideration of the neurochemical intermediates that may contribute to the observed electrophysiological changes. In this structured narrative review, primarily focused on epilepsy, we examine how VNS-related EEG findings can be interpreted in light of noradrenergic, serotonergic, cholinergic, GABAergic, and neurotrophic mechanisms. We also discuss methodological challenges in the analysis of periodic and aperiodic EEG components and outline how machine learning approaches and adaptive closed-loop neuromodulation strategies may support the development of clinically useful VNS biomarkers.
Editorial: Mechanisms and interventions for enhancing cognitive reserve in aging populations
Experimental investigation of sharp-tip microwire-based brain electrode buckling during implantation through dura and pia mater
IntroductionWide use of miniaturized and flexible microwire electrodes faces challenges of wire buckling against the brain membrane layers. The field lacks quantitative understanding of such buckling phenomena, especially on the effective length factor, which is required to determine the wire’s critical buckling load.MethodsThis study presents an experimental investigation into the buckling behavior of tungsten microwire electrodes during implantation through dura and pia mater layers using a validated multilayer brain-mimicking phantom. Microwires with three diameters (25.4, 50.8, and 76.2 µm) and different tip geometries—including blunt, beveled, and electrochemically (conical) sharpened profiles—were evaluated under controlled axial insertion. Critical buckling length, insertion outcomes (buckled/penetrated), and rupture/buckling force were quantified across the experimental dataset. Buckling behavior was analyzed using the Euler column framework with experimentally estimated effective length factors (Kˆ) to represent each unique membrane-wire tip boundary interaction.ResultsResults indicated that wire diameter strongly influences buckling resistance, with larger diameters yielding quartic (fourth order) higher critical buckling load of the electrode, whereas the corresponding membrane rupture force only increases linearly with the diameter. But smaller microwires tend to anchor better against the brain membrane, generating a more stable wire-membrane interface closer to the ideal pin end condition. Tip geometry also significantly affected rupture force and insertion stability; conical tips dramatically reduced the membrane rupture force with less variance. In general, tip sharpening choice for small microwires should focus on optimizing tips anchoring mechanism and minimizing rupture force uncertainty introduced by tip asymmetry while thick microwires mainly benefit from membrane rupture force reduction. For theoretical prediction of a microwire electrode’s critical buckling load based on Euler’s buckling equation, unlike conventional fixed-pinned assumption (K = 0.7), experimentally measured effective length factors ranged from approximately 0.72 – 0.82.DiscussionDesigning with ≈ 0.8 provides a conservative estimate that may reduce the risk of buckling under membrane penetration conditions compared to the commonly assumed fixed-pinned value of 0.7. These findings provide quantitative design guidance for optimizing microwire geometry and offer a validated benchtop framework for predicting buckling-limited insertion performance in neural interface applications.
Oxygen extraction fraction is differentially associated with pathological biomarkers in Alzheimer’s disease and non-Alzheimer’s dementias
IntroductionWe aimed to understand the pathophysiological differences between 16 Alzheimer’s disease (AD) and 15 non-AD dementia patients by quantifying oxygen extraction fraction (OEF) in cortical (CGM) and deep gray matter (DGM) regions.MethodsTo achieve this, we used a novel MRI-based OEF mapping technique, QQ, which estimates OEF from routine multi-echo gradient echo data. Multiple linear regression analyses were performed to compare the associations between OEF and white matter hyperintensities (WMH) or cognitive impairment (measured by Montreal Cognitive Assessment (MoCA) between the two groups.ResultsIn the AD and non-AD group, OEF showed negative associations with WMH in DGM and positive associations with MoCA in DGM and CGM.DiscussionOur study suggests that QQ is a promising tool for differentiating between AD and non-AD dementias, by revealing abnormalities in tissue oxygen usage and their relationships to microvascular changes and cognitive impairment.
Functional connectivity changes in the thalamocortical network due to neck pain and the multiscale regulatory effects of acupuncture: a cross-scale multi-omics neuroimaging study
BackgroundNeck pain correlates with multiscale brain abnormalities, but cross-scale mechanisms of acupuncture analgesia are unclear. This study aimed to: (1) Explore differential modulation of thalamic functional networks by verum vs. sham acupuncture; (2) Examine associations between functional connectivity changes and micro gene expression to unravel its multiscale mechanisms.MethodsA total of 130 participants were initially enrolled, and 100 eligible neck pain patients were randomized 1:1 to the verum (n = 50) or sham (n = 50) acupuncture groups. Finally, 49 patients in each group were included for the final analysis due to one case of exclusion in each group, with treatment administered twice a week for 2 weeks. Visual Analog Scale (VAS), resting-state functional magnetic resonance imaging (fMRI), and Allen Human Brain Atlas (AHBA) transcriptome data were analyzed via Partial Least Squares (PLS) regression.ResultsBoth groups showed reduced post-treatment VAS (p < 0.001), with the verum group exhibiting a superior effect (Z = −6.877, p < 0.001). Neuroimaging revealed that verum acupuncture (VA) specifically induced significant decreases in functional connectivity (FC) between the right thalamus and left anterior cingulate cortex (T = −4.498) as well as between the right thalamus and right Rolandic operculum (T = −4.532, voxel-level p < 0.01, cluster-level p < 0.05), an effect absent in the sham acupuncture group (SA). Gene- FC association analysis indicated that PLS2 component explained 39.83% of FC variance (Pspin: permutation test p < 0.05), with weight genes showing significant spatial correlation to connectivity changes (r = 0.445, Pspin = 0.0011). A total of 809 genes were enriched in the innate immune response and phosphorylation regulation pathways, whereas 1,222 genes were enriched in the GABA-ergic synapse and synaptic membrane-related pathways.ConclusionVA relieves pain via modulating thalamus-anterior cingulate cortex networks, involving immune-inflammation and neural inhibition, providing first multi-scale validation integrating neuroimaging and transcriptomics.Clinical trial registrationThis trial was registered with the International Traditional Medicine Clinical Trial Registry (registration number: ITMCTR2023000001) prior to participant enrollment.
Quality, reliability, and transparency of late-life depression videos on Chinese social media: a cross-sectional study of Douyin, Rednotes, and BiliBili
BackgroundLate-life depression is common in older adults and is often under-recognized. Short-video platforms have become a major source of mental health information. However, content quality and transparency remain uncertain.MethodsWe conducted a cross-sectional assessment of highly viewed videos on late-life depression on three Chinese platforms. We searched each platform using the keyword in Chinese “Late-life depression”. We selected the top 200 videos by view count on Douyin, Rednotes (Xiaohongshu), and BiliBili. After exclusions, 562 videos were included (Douyin, n=188; Rednotes, n=188; BiliBili, n=186). Two medically trained raters scored videos using the Global Quality Score (GQS), modified DISCERN (mDISCERN), and JAMA benchmark criteria. We also coded content categories and creator types. We assessed platform differences using non-parametric tests. We examined associations between a limited engagement proxy, defined as the comment-to-view ratio, and quality scores using Spearman correlation.ResultsVideo duration differed across platforms (p<0.001). Engagement indicators were higher on Douyin and Rednotes than on BiliBili. Symptoms were the most common topic on all platforms. Prevention and intervention ranked second on Douyin and Rednotes. On BiliBili, causes and case-based analysis were also common. Overall quality was moderate. Mean GQS ranged from 2.96 to 3.05. Transparency was limited. Mean JAMA ranged from 1.91 to 2.04. Reliability was slightly higher on BiliBili based on mDISCERN. Creator type was strongly associated with scores. Expert and institutional videos scored higher than general and marketing-oriented accounts. Correlations between visible audience interaction and quality were weak.ConclusionHighly viewed late-life depression videos on major Chinese platforms show moderate quality and limited transparency. Exposure does not reliably signal higher-quality information. Platforms and health authorities should strengthen source disclosure and promote evidence-based content from qualified creators.
Safety and preliminary efficacy of Aurora: a pilot, non-randomized clinical trial of a culturally adapted digital cognitive behavioral therapy intervention for anxiety and depression in Mexico
Background/objectiveAnxiety and depressive disorders are leading causes of disability worldwide, and access to evidence-based psychological treatment remains limited in many middle-income countries. Digital cognitive–behavioral therapy (CBT) interventions have emerged as scalable tools to address this treatment gap, yet few have undergone clinical evaluation in Latin American populations. This study aimed to assess the safety and preliminary efficacy of Aurora, a Spanish-language, culturally adapted digital CBT program, when used as an adjunct to pharmacotherapy in adults with generalized anxiety disorder.MethodsIn a multicenter, open-label, non-randomized pilot study, 34 adults diagnosed with generalized anxiety disorder receiving stable pharmacological treatment were assigned through pragmatic, convenience-based allocation either to an experimental group (Aurora plus medication; n = 24) or to a control group receiving medication alone (n = 10). The sample had a mean age of 39.85 ± 12.88 years, with a predominance of women (22/34). Participants were followed for 12 weeks with assessments at baseline and weeks 4, 8, and 12. Clinical outcomes included anxiety severity measured by the Generalized Anxiety Disorder-7 (GAD-7), pathological worry assessed by the Penn State Worry Questionnaire (PSWQ), and depressive symptoms evaluated using the Patient Health Questionnaire-9 (PHQ-9). Safety was monitored through structured adverse-event reporting. Statistical analyses included linear mixed-effects models for longitudinal outcomes, ordinal logistic regression for severity transitions, and negative binomial regression and Fisher’s exact test for adverse events, with false discovery rate correction applied where appropriate.ResultsAurora demonstrated a favorable safety profile, with no serious adverse events and comparable adverse-event incidence between groups under structured clinical monitoring at weeks 4, 8, and 12. Anxiety symptoms (GAD-7) showed a significant effect of time (F3,96 = 169.65; p < 0.001), indicating reductions across both groups. Pathological worry (PSWQ) demonstrated significant group (F1,31.12 = 6.96; p = 0.013) and group × time interaction effects (F3,93.4 = 7.86; p < 0.001), with greater reductions in the Aurora group, particularly at weeks 8 and 12. At week 12, ordinal analyses indicated higher odds of lower worry severity in the intervention group (β = 2.53; p = 0.004; OR = 12.5). Depressive symptoms decreased similarly in both groups. Positive effect increased progressively across intervention modules, and module-embedded cognitive measures of anxiety and depression showed significant reductions over time.ConclusionThis pilot study provides preliminary, hypothesis-generating evidence that a culturally adapted digital CBT intervention can be safely integrated with pharmacotherapy and may be associated with enhanced improvements in anxiety-related outcomes, particularly pathological worry, in a Mexican clinical population. However, the non-randomized design, small sample size, and baseline imbalances limit causal inference and generalizability, and findings should be interpreted with caution. Larger randomized controlled trials are needed to confirm efficacy, determine long-term clinical impact, and guide the implementation of digital therapeutics in Latin American mental health systems.