Brain · AI · Top Stories
A curated, continuously updated feed of the most consequential stories at the intersection of artificial intelligence and human cognition. Most items link straight to the social media post where the news is breaking. Tap a headline to read the thread, or submit a story.
A specialized brain imaging AI trained on 5.24 million hospital MRI and CT volumes outperformed GPT-5 by 21.4 percentage points on identifying critical radiology findings in a real-world hospital trial, highlighting the advantage of domain-specific models over general-purpose LLMs in medical imaging.
A commentary examines Meta's AI research that decodes brain activity to reconstruct sentences typed by participants. The piece notes the achievement is significant but distinct from true mind-reading, as the system reconstructs known typed input rather than freely reading thoughts.
Meta's FAIR team demonstrated Brain2Qwerty v2, a system that translates brain activity into typed sentences using non-invasive recordings rather than surgical implants. The work signals progress toward brain-computer interfaces that do not require invasive procedures.
Researchers combined recordings from individual neurons with artificial-intelligence techniques to identify specific brain cells involved in speech production. The findings could advance understanding of how the brain generates language and inform future speech-restoring neurotechnology.
A report describes advanced brain-interfacing technology aimed at restoring sensory functions including touch and vision, integrating AI-driven approaches. Such neural interfaces could offer new avenues for people with sensory impairments.
Scientists studied the brain circuitry linking visual perception and cognition, using AI models to complement brain scans that lack single-cell resolution. The work sheds light on how neural activity supports both seeing and thinking.
Meta unveiled Brain2Qwerty, an AI-powered brain-computer interface that translates brain activity into typed text. The system is aimed at aiding communication for people with neurological disorders.
Researchers report using a brain-computer interface headset to detect covert consciousness in patients with brain injuries who appear unresponsive. The approach could help identify awareness that standard bedside assessments miss, informing diagnosis and care decisions.
Meta AI releases Brain2Qwerty v2, the highest-performing end-to-end pipeline capable of real-time sentence decoding from non-invasive brain recordings, approaching levels of accuracy previously exclusive to techniques that require brain surgery.
Researchers studying brain activity under anesthesia found patterns that challenge current understanding of consciousness, noting that the brain's predictive behavior resembles how large language models operate. The findings could inform both neuroscience theories of consciousness and AI development.
The piece argues that the first foundation models of the brain emerged only in neuroscience subfields that developed standards for integrating data. It contends AI cannot substitute for the field-wide work of making datasets compatible, a prerequisite for building useful brain models.
A video report features a brain-computer interface that enables a man with ALS to communicate, demonstrating how implanted neural devices can restore speech for people with paralysis.
A report surveys the rapid emergence of brain-computer interfaces, AI-powered smart glasses, and robotic prosthetic arms, noting that technologies once confined to cyberpunk fiction are increasingly entering real-world use. It highlights the convergence of neurotech and AI in human augmentation.
Europe's first exascale supercomputer, JUPITER, built on NVIDIA Grace Hopper Superchips, is being used for large-scale science including detailed brain mapping alongside climate modeling and 6G AI research. Exascale computing power enables higher-resolution neuroscience simulations and analyses.
Researchers at the University of Oxford developed an AI tool that analyzes imaging data—including brain MRI, blood vessels, kidneys, liver and body composition—to reveal organ damage caused by high blood pressure. The approach could enable earlier detection of hypertension-related harm.
An opinion piece argues that efforts by Eleos AI Research and Anthropic to probe consciousness in large language models are unconvincing, contending that sentience in the brain is mechanized by neurons and their electrical and chemical signals\u2014processes LLMs do not replicate.
A first-person account of participating in a brain experiment investigating how biological neural systems process information. Researchers aim to apply principles of biological computation to make artificial intelligence less energy-intensive.
Scientists are aligning the internal representations of pretrained large language models with brain activity data, an approach called brain-tuning, to make AI models function more like biological brains. The goal is to use these models as useful 'model organisms' for studying neural computation.
Insilico Medicine and SK Biopharmaceuticals signed a heavily backloaded deal worth up to $2.5 billion to use Insilico's AI platform to discover drugs for neurological disorders. The partnership highlights growing investment in AI-driven discovery targeting brain and nervous system diseases.
ChristianaCare researchers published work advancing an AI-guided radiation therapy approach for glioblastoma, an aggressive brain cancer. The method aims to predict and more precisely target tumor regions, potentially improving treatment outcomes.
OpenAI's o3 Deep Research model teamed with clinicians at Boston Children's Hospital and Harvard to re-examine 376 rare, undiagnosed pediatric cases. The model proposed diagnostic hypotheses linking symptoms, genetic variants, and the medical literature; doctors validated them against ACMG/AMP clinical standards, confirming 18 brand-new diagnoses across neurodevelopmental, neuromuscular, and psychiatric conditions. The work was published in NEJM AI.
Image-AI company Midjourney unveiled a medical division built around a conceptual full-body scanner that images in 60 seconds with no radiation or contrast dye, aiming to replace a large share of MRIs. A ring of 8,960 ultrasonic transducers claims picometer-scale resolution at 17 GB/s. The bootstrapped company says it will deploy 50,000 units over six years and open a San Francisco wellness spa in 2027, partnered with Butterfly Health.
Samsung's foundry division has secured its first-ever order from Neuralink, taping out test silicon for the company's fourth-generation brain-computer interface implant, internally codenamed "O1." The chip is built on Samsung's energy-efficient 4-nanometer process — a node that shrinks the implant, speeds up neural-signal decoding, and cuts power draw, all critical to how long a device can safely live in the cortex. Samsung kicked off the project late last year; first samples are due in the first half of 2027, with mass production possible by the second half if testing holds up.
Curated for educational purposes. Story summaries are condensed from public reporting; tap through to the original source for the full context.
Submit a story
This feed is hand-curated as the Brain-AI story develops. Want your news here, or have a tip? Reach out and we'll take a look.
Submit a story
Spotted brain-AI news worth featuring? Send it over — every submission is reviewed before it goes live.