LeWorldModel strengthens Yann LeCun’s long standing argument that latent predictive world models may be far more efficient than brute force generative approaches for learning actionable structure from the world. It doesn’t kill gen AI, but makes alternative harder to ignore doi.org/10.48550/arX...

I have co-authored with my European colleagues a study on international non-proprietary name (INN) - Europe has the legal and technical foundation to solve this. - A highly relevant paper for anyone working in digital health, interoperability, or medication safety. academic.oup.com/eurpub/artic...

🚨 Final Call -Patient-Powered Digital Health 2026 June 22–24, 2026 at Harvard & BIDMC 1. Research Track - For validated work from researchers, clinicians, and industry 2. Innovation Track - For early-stage and high-potential digital health solutions ⏰ Due: 4/17/26 11:59 PM EST lnkd.in/evTNrmGQ

Final call for proposals — Patient-Powered Digital Health 2026, June 22–24, Harvard/BIDMC. Keynote: Dr. Jag Singh, HMS/MGH. Evidence-based, patient co-designed work only. Deadline: April 17 www.dcinetwork.org/patients2026

Explore update scope of IJMI - Health information systems - Clinical decision support systems and AI - Education and workforce - Evidence synthesis - Safety and policy IJMI introduces There is new article type: Implementation - real-world informatics impact. www.sciencedirect.com/journal/inte...

Anthropic accidentally leaked ~512,000 lines of proprietary TypeScript source code for its Claude Code CLI tool. The leaked code is now permanently available via mirrors, decentralized backups, and the open-source rewrites. Claw-code repo (Sigrid Jin’s Python rewrite github.com/instructkr/c...

I invite you to Patient-Powered Digital Health 2026 www.dcinetwork.org/patients2026 Peer-reviewed scientific presentations with PubMed-indexed proceedings, panels, case studies, workshops, themes: AI-powered patient engagement, patient co-design, trustworthy and ethical AI, evaluation of impact.

How to optimize your google search cardcatalogforlife.substack.com/p/google-has...

MIT’s Neural Thickets argues that large pretrained models already contain many nearby task experts. Their simple RandOpt method samples random weight nudges, keeps the best, and ensembles them, rivaling heavier post training on some tasks. #AI arxiv.org/abs/2603.12228

Google’s release of multimodal embedding models represents a significant step in unifying diverse data types—text, images, and video—into a single high-dimensional vector space. You can integrate unstructured data with imaging and video to enhance precision medicine blog.google/innovation-a...

Check out x.com/yuriquintana...

Google drops another #AI #AcademicAI strong research tool: PaperBanana. An agentic system that generates publication ready academic figures directly from methodology descriptions. No manual diagramming or Figma needed. Just your method section and caption. arxiv.org/abs/2601.23265

#AIPatents #HealthcareInnovation #LegalChallenges PTAB rulings emphasize disclosure hurdles for AI patents in healthcare due to neural network opacity. www.reuters.com/legal/legali...

#AIGovernance #ClinicalAI #FDARegulation New framework proposes treating generative AI as novel intelligence for clinical oversight, modeled on human certifications. jamanetwork.com/journals/jam...

#AIDrugDiscovery #CKDAnemia #ClinicalTrials AI-driven CKD anemia candidate ISM4808 completes first human dosing in Phase I trial, advancing from discovery to clinic rapidly. markets.ft.com/data/announc...

#ClinicalTrials #AIMatching #HealthcareAI DocTr AI framework matches clinicians to trials using multimodal data, improving accuracy and fairness. www.nature.com/articles/s44...

#AIHealthcare #LifeSciences #NVIDIARport NVIDIA report shows AI advancing radiology, drug discovery, and digital twins in healthcare. x.com/nvidia/statu...

#PsychotherapyAI #MentalHealthTech #AdoptionBarriers Study identifies facilitators and barriers for AI in psychotherapy, emphasizing training and privacy. www.nature.com/articles/s44...

#AISafety - LLMs lack morality even with guardrails. New paper finds that GPT can show unexpected gender bias, not only in stereotypes but also in moral dilemma judgments. Important reminder that AI fairness issues can be subtle, implicit, and context dependent. www.sciencedirect.com/science/arti...

in Nature: Open access Evo 2, a 40B-param foundation model trained on 9T DNA bases, predicts variants & designs genomes across life domains. Zero-shot pathogenicity, interpretable features, chromatin control—open-sourced for biodesign! #AIinBiology #Genomics www.nature.com/articles/s41...

#IJMI open access paper: SMART on FHIR improved postpartum VTE risk assessment in randomized crossover simulation. Semi-automated app increased accurate risk scores (63%→87%), improved prophylaxis recommendations (79%→91%), reduced time and data entry burden www.sciencedirect.com/science/arti...

New Nature Medicine analysis maps 4,609 clinical LLM evaluation studies (2022–Sep 2025). Only 1,048 used real patient data and just 19 were RCTs; most were simulated or exam style. Evidence remains thin for adoption. #AIGovernance doi.org/10.1038/s415...

Significant shift in corporate strategy & investor sentiment, "pilot" now is poorly viewed, companies have run small-scale tests that never scale into meaningful business value. Earnings calls now use "proof of concept" emphasizing a path toward commitment. Where is ROI? www.wsj.com/tech/ai/pilo...

Canada stands at a pivotal strategic inflection point in AI. A national AI strategy consultation drew over 11,300 inputs from Canadians—the largest such engagement in Innovation, Science and Economic Development Canada's history, revealing a strong consensus views. www.linkedin.com/pulse/canada...

Literacy screenings, not typically conducted in medical settings, are now being done at pediatric care centers and providing support resources, recognizing reading’s importance for a child’s development. www.cnn.com/2026/02/26/h...

Oura launches a proprietary AI model focused on women’s health techcrunch.com/2026/02/24/o...

FDA new policy for new treatments for rare diseases based on evidence for a "plausible mechanism" for treatment would work. aims to speed new technologies like gene-editing that are so rare that it would be difficult if not impossible to conduct a traditional study first. www.npr.org/2026/02/23/n...

GitNexus turns any GitHub repo into an in browser interactive knowledge graph plus AI chat. It runs a 4 pass pipeline (structure, Tree sitter AST parsing, dependency resolution, call graph) with zero server and no code leaving your device. 100% free and open source github.com/abhigyanpatw...

iKraph (Nature Machine Intelligence, 2025) builds a large-scale biomedical knowledge graph from all PubMed abstracts using a LitCoin-winning NLP pipeline. It identified ~1,200 COVID drug candidates early, with one-third later validated. www.nature.com/articles/s42...

A new “vectorless RAG” approach, PageIndex, replaces chunking & embeddings with a hierarchical table-of-contents tree. model navigates sections step-by-step; reports 98.7% accuracy on FinanceBench, outperforming traditional RAG. Tradeoff: higher cost & latency. #AIResearch pageindex.ai/blog/Mafin2.5

Anthropic introduces a Persona Selection Model: AI assistants do not have fixed personalities but probabilistically select human like personas shaped by training data, prompts, and alignment tuning. A new lens on AI behavior and safety. alignment.anthropic.com/2026/psm | #AIPersonhood #AIGoverance

Grok, Gemini, Claude and ChatGPT are all still in the race. The leaderboard keeps changing weekly. The open source may have a big impact. Over the next year it will clearer which strategy(Cloud vs On premise ) & LLM makes sense for which use case, and which company is the first to run out of money.

Google DeepMind paper on Pragmatic #AIpersonhood -“person” as a configurable bundle of obligations, not a metaphysical status. Highlights risks (dark patterns, dehumanization) and sketches identity plus sanction based accountability architectures arxiv.org/abs/2510.26396

Recursive self improvement is the real test of an AI explosion. Darwin Gödel Machine, AlphaEvolve, and Meta’s self rewarding models show rapid gains, but mostly narrow optimization, not self rewriting intelligence. Acceleration is real, takeoff is not. arxiv.org/abs/2505.22954 #AI

Video by Marine Sgt Maj Carlos Ruiz urges Marines to support one another &seek help when struggling. He stressed asking for help is strength & there is no wrong door to walk through because others stand ready to assist. Leadership means ensuring your team supported taskandpurpose.com/military-lif...

Google Gemini 3.1 Pro just released has passed other LLMs on Humannity's Last Exam. blog.google/innovation-a...

Progress!

New 100 page paper on Agentic Reasoning for LLMs from researchers at U Illinois, Meta, Amazon, Google DeepMind, UCSD, and Yale argues true AI autonomy requires iterative planning, memory, tool use, and feedback loops beyond chain of thought. #AIResearch arxiv.org/pdf/2601.12538

AI firms face same dilemma as social platforms: when do private user conversations signal a public safety risk? WSJ report details internal concerns raised months before a Canada shooting suspect drew attention. How to balance Privacy vs public protection? #AIGovernance www.wsj.com/us-news/law/...

Utah’s AI sandbox is piloting AI prescription renewals. Doctronic may renew non controlled maintenance refills previously prescribed, with identity checks and clinician escalation. Findings may inform scope. #HealthAI commerce.utah.gov/ai/agreement...

FDA TEMPO: devices may be used in CMS ACCESS pre-authorization; RWE supports clearance. #DigitalHealth www.fda.gov/news-events/...

CMS ACCESS shifts Medicare chronic care from visit billing to outcome aligned recurring payments, rewarding measurable improvement and enabling scalable, tech supported care beyond the clinic. #ACCESSModel www.cms.gov/priorities/i...

Amid Lawsuits, OpenAI Says It Will Retire “Reckless” Model Linked to Deaths futurism.com/artificial-i...

OpenAI Retires GPT-4o as Lawsuits Mount openai.com/index/retiri...

Osler (2026) argues that the real risk is not just AI “hallucinations,” but distributed delusions that emerge when users rely on chatbots for memory and meaning making. Affirming dialogue can entrench false beliefs, particularly under social isolation. #AIGovernance link.springer.com/article/10.1...

UK PM Keir Starmer will extend Online Safety Act rules to AI chatbots after the Grok controversy, empowering Ofcom to fine or ban services that generate illegal content harmful to children. A major shift toward AI accountability. #AIGovernance www.theguardian.com/technology/2...

#AICardiology #HealthEconomics #ClinicalAI npj Digital Medicine study simulates AI decision support for coronary revascularization on 25k+ patients, showing cost savings & better outcomes in Alberta data. www.nature.com/articles/s41...

Feb 2026 1. OpenAI GPT-5.2 dominates mathematical reasoning 2. Google Gemini 3 Pro leads in multimodal & visual reasoning 3. Anthropic Claude Opus 4.6 strong scientific reasoning 4. DeepSeek-R1 are competitive, particularly in math reasoning 5. Most now exceed 90%+ on PhD-level science questions

LLMs Benchmarks LM Council lmcouncil.ai/benchmarks Epoch AI Benchmarks with FrontierMath epoch.ai/benchmarks LLM-Stats Benchmarks (GPQA, MMLU, AIME) llm-stats.com/benchmarks LiveBench Contamination-free live benchmark livebench.ai Stanford HAI AI Index 2025 AI Index Report hai.stanford.edu/ai-index

MiniMax Released 2/12/26, on Hugging Face and GitHub, M2.5 scores 80.2% on SWE-Bench Verified, matching Anthropic's Claude Opus 4.6, while running 37% faster than its predecessor. API costs $1 per hour, 10-20% of rivals, self-hosting is free, integrate it with OpenClaw www.minimax.io/news/minimax...