Artificial Intelligence

Every cloud provider faces the same AI infrastructure challenge: chips need to be positioned close together to exchange data quickly, but they generate intense heat, creating unprecedented cooling demands. We needed a strategic solution that allowed us to use our existing air-cooled data centers to do liquid cooling without waiting for new construction. And it needed to be rapidly deployed so we could bring customers these powerful AI capabilities while we transition towards facility-level liquid cooling. Think of a home where only one sunny room needs AC, while the rest stays naturally cool – that’s what we wanted to achieve, allowing us to efficiently land both liquid and air-cooled racks in the same facilities with complete flexibility. The available options weren't great. Either we could wait to build specialized liquid-cooled facilities or adopt off-the-shelf solutions that didn't scale or meet our unique needs. Neither worked for our customers, so we did what we often do at Amazon… we invented our own solution. Our teams designed and delivered our In-Row Heat Exchanger (IRHX), which uses a direct-to-chip approach with a "cold plate" on the chips. The liquid runs through this sealed plate in a closed loop, continuously removing heat without increasing water use. This enables us to support traditional workloads and demanding AI applications in the same facilities. By 2026, our liquid-cooled capacity will grow to over 20% of our ML capacity, which is at multi-gigawatt scale today. While liquid cooling technology itself isn't unique, our approach was. Creating something this effective that could be deployed across our 120 Availability Zones in 38 Regions was significant. Because this solution didn't exist in the market, we developed a system that enables greater liquid cooling capacity with a smaller physical footprint, while maintaining flexibility and efficiency. Our IRHX can support a wide range of racks requiring liquid cooling, uses 9% less water than fully-air cooled sites, and offers a 20% improvement in power efficiency compared to off-the-shelf solutions. And because we invented it in-house, we can deploy it within months in any of our data centers, creating a flexible foundation to serve our customers for decades to come. Reimagining and innovating at scale has been something Amazon has done for a long time and one of the reasons we’ve been the leader in technology infrastructure and data center invention, sustainability, and resilience. We're not done… there's still so much more to invent for customers.

The Agentic Enterprise is driving profound change across every industry, but nowhere are the stakes higher than in healthcare. There is an incredible opportunity to elevate the work of healthcare professionals and deliver stronger care for patients around the world. In an essay for TIME, Murali Doraiswamy, professor of medicine at Duke University, and I discuss how AI is revolutionizing medicine, including: • Flagging subtle abnormalities in scans and slides that a human eye might miss. • Speeding up the discovery of drugs and drug targets. • Providing patients faster and more personalized support, from scheduling to flagging side effects But we’ve also seen that over-reliance on AI can lead to “deskilling” — in which medical professionals become less effective. That underscores the importance of approaches that keep humans at the center, such as the Intelligent Choice Architecture (ICA), where AI systems don’t make decisions but nudge providers to take a second look at results, weigh alternatives, and stay actively engaged in the process. The future of work is humans and AI agents working together. If we commit to designing systems that sharpen our abilities, we can combine the promise of AI with the critical thinking, compassion, and real-world judgment that only humans bring. https://lnkd.in/gqkTUfb6

Three AI recruiters look at the same 109 CVs. They agree only 14% of the time. That’s not the start of a joke. And that's not efficiency. That’s what I call 'Rank Roulette'. When I tested ChatGPT, Gemini and Grok against the same job spec and anonymised CV set, here’s what happened: • 14% overlap in shortlists → Four times out of five, the models disagreed. • ±2.5 places volatility → Yesterday’s #2 became today’s #5. • 55% of CVs never surfaced → Candidates vanished with no audit trail. • 96% recycled rationales → Fluent, but shallow logic. We’re told by vendors and in-house 'tinkerers' that LLMs can “shortlist in seconds”. The truth: they behave more like over-confident interns - smooth on the surface, but shockingly inconsistent. And the worst part? It’s not even random. In a follow-up piece, I explored why this happens: a technical quirk called batch non-determinism. In plain English: your candidate’s fate changes depending on what else the server was processing at that moment. Until volatility is tamed, hands-off AI screening with LLMs is more than risky. It’s completely unexplainable, indefensible and a governance nightmare. Go to the comments for 👉 Full research 👉 Follow-up on why AI recruiters play favourites

Microsoft just released a 35-page report on medical AI - and it’s a reality check for healthcare. The paper, “The Illusion of Readiness”, tested six of the most popular models (OpenAI, Gemini, etc)… across six multimodal medical benchmarks. And the verdict? The models scored high on medical exams. But they’re not even close to being real-world ready. Here’s what the stress tests revealed: ▶ 1. Shortcut learning Models often answered correctly even when key information, like medical images, was removed. They weren’t reasoning - they were exploiting statistical shortcuts. That means benchmark wins may hide shallow understanding. ▶ 2. Fragile under small changes Making small tweaks caused big swings in predictions. This fragility shows how unreliable model reasoning becomes under stress. In visual substitution tests, accuracy dropped from 83% to 52% when images were swapped - exposing shallow visual–answer pairings. ▶ 3. Fabricated reasoning Models produced confident, step-by-step medical explanations - but many were medically unsound… or entirely fabricated. Convincing to the eye, dangerous in practice. And more importantly, healthcare isn’t a multiple-choice exam. It’s uncertainty, incomplete data, and high stakes. So Microsoft’s team calls for new standards: - Stress tests that expose fragility - Clinician-guided guidelines that profile benchmarks - Evaluation of robustness and trustworthiness - not just leaderboard scores The takeaway is simple: Medical AI may ace tests today. But until it proves reliable under stress, it’s not ready for the clinic. When do you think popular LLMs will be clinic-ready? #entrepreneurship #healthtech #AI

My biggest takeaways from Ethan Smith on how to win at AEO (i.e. get ChatGPT to recommend your product): 1. Being mentioned most often beats ranking first. In Google, the #1 blue link wins. In ChatGPT, the answer summarizes multiple sources—so appearing in five citations beats ranking #1 in one. Ethan’s strategy: get mentioned on Reddit, YouTube, blogs, and affiliates. Volume of mentions matters more than any single placement. 2. LLM traffic converts 6x better than Google search traffic. Webflow saw this dramatic difference because users who come through AI assistants have built up much more intent through conversation and follow-up questions, making them highly qualified leads. 3. Early-stage startups can win at AEO immediately, unlike with SEO. Traditional SEO requires years of domain authority. But a brand-new Y Combinator company mentioned in a Reddit thread today can show up in ChatGPT tomorrow. The playing field is finally level. 4. The long tail of AEO is 4x bigger than SEO. People ask ChatGPT questions with 25 or more words (vs. 6 in Google). Ethan found gold in queries like “Which meeting transcription tool integrates with Looker via Zapier to BigQuery?”—questions that never existed in search but are perfect for AI. Own these micro-niches. 5. Reddit is proving to be the kingmaker for AI visibility. ChatGPT trusts Reddit because the community polices spam better than any algorithm. Ethan’s exact playbook: make one real account, say who you are and where you work, give genuinely helpful answers. Five good comments can transform your visibility. No automation, no fake accounts—just be helpful. 6. YouTube videos for “boring” B2B terms are a gold mine for AEO. Nobody makes videos about “AI-powered payment processing APIs”—which is exactly why you should. While everyone fights over “best CRM software,” the high-value, zero-competition long tail is wide open in video. 7. Your help center is now a growth channel. All those “Does your product do X?” questions flooding ChatGPT can be answered by help-center pages. Move them from subdomain to subdirectory, cross-link aggressively, and cover every feature question. Ethan calls this the most underutilized opportunity in AEO. 8. January 2025 was the inflection point in AEO growth. That’s when ChatGPT made answers more clickable (maps, shopping cards, citations) and adoption exploded. Webflow went from near zero to 8% of signups from AI. This channel is accelerating faster than any Ethan’s seen in 18 years. 9. The AEO playbook: (1) Find questions from competitor paid search data, (2) set up answer tracking, (3) see who’s showing up as citations, (4) create landing pages answering all follow-up questions, (5) get mentioned offsite via Reddit/YouTube/affiliates, (6) run controlled experiments, (7) build a dedicated team. This exact process is driving real results at scale.

Something VERY cool just happened in California and… it could be the future of energy.   On July 29, just as the sun was setting, California’s electric grid was reaching peak demand.   However, instead of ramping up fossil fuel resources, the California Independent System Operator (CAISO) and local utilities decided to lean on a network of thousands of home batteries.   More than 100,000 residential battery systems (made up primarily by Sunrun and Tesla customers) delivered about 535 megawatts of power to California’s grid right as demand peaked, visibly reducing net load (as shown in the graphic).   Now, this may not seem like a lot but 535 megawatts is enough to power more than half of the city of San Francisco and that can make all the difference when a grid is under stress.   This is what’s called a Virtual Power Plant or VPP. It’s a network of distributed energy resources that grid operators can call on in an emergency to provide greater resilience to our energy systems. Homeowners are compensated for the dispatch, grid operators are given another tool for reliability, and ratepayers are saved from instability. It’s a win-win-win.   Now, this was just a test to prepare for other need-based dispatches during heat waves in August and September. But it’ historic.   As homeowners add more solar and storage resources, the impact of these dispatch events will become even more profound and even more necessary. This was the second time this summer that VPPs have been dispatched in California and I expect to see even more as this technology improves.   Shout out to Sunrun, Tesla, and all companies who participated. Keep up the great work.

We’ve all heard about AI’s potential to boost productivity. But what truly matters to me is whether it’s making work better for the people who show up every day. At Cisco, our People Intelligence team, in collaboration with IT, has been exploring this very topic, and the findings are fascinating. Here are five key insights from our research that leaders should take seriously: 1. Leaders are key to adoption. At Cisco, employees are 2x more likely to use AI if their direct leader uses it. 2. Generic AI training doesn’t work. Role-specific, practical training accelerates AI use. 3. Confidence gaps exist among senior leaders. Directors at Cisco often feel less confident with AI than mid-level employees, underscoring the need for tailored support at all levels. 4. Employee autonomy fuels adoption. Hybrid work environments are powerful accelerators for AI adoption, while mandates can hinder it. Employees who voluntarily go to the office are more likely to use AI, while those who are required to work on-site have lower adoption. 5. AI use is linked to employee well-being, but the relationship is complex, with both benefits and trade-offs that require thoughtful navigation. This is just the beginning. Next, we’re looking at how AI is transforming the way teams operate. For now, one thing is clear, employees who use AI aren’t just more productive. They’re also more engaged, better aligned with company strategy, and empowered to focus on meaningful work. #AIAdoption #EmployeeExperience #FutureOfWork

I taught myself machine learning > 10 years ago. If I had to start again today, I wouldn’t touch models, LLMs, or agents first, as many AI experts suggest. I'd start with the math and the code. Ugly truth: 90% of people skip the foundations, then wonder why everything feels like magic or falls apart in production. If you want to be different, actually understand ML, not just copy-paste, this is the roadmap I'd follow: Start with fundamentals: Because no matter how fast LLMs or GenAI evolve, your math, code, and logic will keep you relevant. Here's what you should focus on: 📐 1. Linear Algebra Learn these core ideas: Vectors, matrices, tensors Matrix multiplication (dot products, broadcasting) Transpose, inverse, rank, determinants Eigenvalues & eigenvectors (especially for PCA & embeddings) Projections and orthogonality ✅ Use NumPy to implement everything yourself → Practice matrix ops, dot products, and visualizing transformations with Matplotlib 🔁 2. Calculus Focus on: Derivatives & partial derivatives Chain rule (for backpropagation in neural nets) Gradient descent Convex functions, minima/maxima ✅ Use SymPy or JAX to visualize and compute derivatives → Plot functions and their gradients to develop deep intuition 🎲 3. Probability You need a solid grip on: Random variables (discrete & continuous) Conditional probability & Bayes' rule Joint & marginal probability The Chain rule Expectation, variance, entropy Common distributions: Bernoulli, Binomial, Gaussian, Poisson Central limit theorem The law of large numbers ✅ Simulate simple probability experiments in Python with NumPy → E.g. simulate sampling from distributions 📊 4. Statistics These are must-know topics: Descriptive stats: mean, median, mode, standard deviation Hypothesis testing: p-values, confidence intervals, t-tests Correlation vs. causation Sampling, bias, and variance Overfitting/underfitting A/B testing basics ✅ Use Pandas & SciPy to explore real datasets → Calculate descriptive stats, create histograms/box plots, run t-tests 🔧 Essential Python libraries to learn early NumPy – for vectorized math and fast array ops Pandas – for loading, cleaning, and analyzing tabular data Matplotlib / Seaborn – for plotting and visualizing distributions, relationships, and trends SymPy – for symbolic math and calculus SciPy – for stats, optimization, and numerical methods Use Jupyter Notebooks(to combine math, code, & visuals in one place) 📚 Best resources to nail the fundamentals: ✅ Machine Learning Foundations Math series (ML Foundations: Linear Algebra, Calculus, Probability, and Statistics)-series of 4 courses that I've created together with LinkedIn learning ✅ Hands-On ML with TensorFlow & Keras book by Aurélien Géron ✅ The Hundred-page Machine Learning Book by Andriy Burkov If you want to become an actual ML engineer, not just someone who watches and copies demos, start here. ♻️ Repost to help others💚

Invisible UX is coming 🔥 And it’s going to change how we design products, forever. For decades, UX design has been about guiding users through an experience. We’ve done that with visible interfaces: Menus. Buttons. Cards. Sliders. We’ve obsessed over layouts, states, and transitions. But with AI, a new kind of interface is emerging: One that’s invisible. One that’s driven by intent, not interaction. Think about it: You used to: → Open Spotify → Scroll through genres → Click into “Focus” → Pick a playlist Now you just say: “Play deep focus music.” No menus. No tapping. No UI. Just intent → output. You used to: → Search on Airbnb → Pick dates, guests, filters → Scroll through 50+ listings Now we’re entering a world where you guide with words: “Find me a cabin near Oslo with a sauna, available next weekend.” So the best UX becomes barely visible. Why does this matter? Because traditional UX gives users options. AI-native UX gives users outcomes. Old UX: “Here are 12 ways to get what you want.” New UX: “Just tell me what you want & we’ll handle the rest.” And this goes way beyond voice or chat. It’s about reducing friction. Designing systems that understand intent. Respond instantly. And get out of the way. The UI isn’t disappearing. It’s mainly dissolving into the background. So what should designers do? Rethink your role. Going forward you’ll not just lay out screens. You’ll design interactions without interfaces. That means: → Understanding how people express goals → Guiding model behavior through prompt architecture → Creating invisible guardrails for trust, speed, and clarity You are basically designing for understanding. The future of UX won’t be seen. It will be felt. Welcome to the age of invisible UX. Ready for it?

McKinsey & Company 𝗮𝗻𝗮𝗹𝘆𝘇𝗲𝗱 𝟭𝟱𝟬+ 𝗲𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝗚𝗲𝗻𝗔𝗜 𝗱𝗲𝗽𝗹𝗼𝘆𝗺𝗲𝗻𝘁𝘀 — 𝗮𝗻𝗱 𝗳𝗼𝘂𝗻𝗱 𝗼𝗻𝗲 𝗰𝗼𝗺𝗺𝗼𝗻 𝘁𝗵𝗿𝗲𝗮𝗱: ⬇️ One-off solutions don’t scale. The most successful projects take a different path: They use open, modular architectures that enable speed, reuse, and control. → Designed for reuse → Able to plug in best-in-class capabilities → Free from vendor lock-in This is the reference architecture McKinsey now recommends — optimized to scale what works while staying compliant. It consists of five core components: ⬇️ 𝟭. 𝗦𝗲𝗹𝗳-𝘀𝗲𝗿𝘃𝗶𝗰𝗲 𝗽𝗼𝗿𝘁𝗮𝗹: → A secure, compliant “pane of glass” where teams can launch, monitor, and manage GenAI apps. → Preapproved patterns, validated capabilities, shared libraries. → Observability and cost controls built-in. 𝟮. 𝗢𝗽𝗲𝗻 𝗮𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲 → Services are modular, reusable, and provider-agnostic. → Core functions like RAG, chunking, or prompt routing are shared across apps. → Infra and policy as code, built to evolve fast. 𝟯. 𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗲𝗱 𝗴𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 𝗴𝘂𝗮𝗿𝗱𝗿𝗮𝗶𝗹𝘀 → Every prompt and response is logged, audited, and cost-attributed. → Hallucination detection, PII filters, bias audits — enforced by default. → LLMs accessed only through a centralized AI gateway. 4. 𝗙𝘂𝗹𝗹-𝘀𝘁𝗮𝗰𝗸 𝗼𝗯𝘀𝗲𝗿𝘃𝗮𝗯𝗶𝗹𝗶𝘁𝘆 → Centralized logging, analytics, and monitoring across all solutions → Built-in lifecycle governance, FinOps, and Responsible AI enforcement → Secure onboarding of use cases and private data controls → Enables policy adherence across infrastructure, models, and apps 5. 𝗣𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻-𝗴𝗿𝗮𝗱𝗲 𝗨𝘀𝗲 𝗖𝗮𝘀𝗲𝘀 → Modular setup for user interface, business logic, and orchestration → Integrated agents, prompt engineering, and model APIs → Guardrails, feedback systems, and observability built into the solution → Delivered through the AI Gateway for consistent compliance and scale The message is clear: If your GenAI program is stuck, don’t look at the LLM. Look at your platform. 𝗜 𝗲𝘅𝗽𝗹𝗼𝗿𝗲 𝘁𝗵𝗲𝘀𝗲 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁𝘀 — 𝗮𝗻𝗱 𝘄𝗵𝗮𝘁 𝘁𝗵𝗲𝘆 𝗺𝗲𝗮𝗻 𝗳𝗼𝗿 𝗿𝗲𝗮𝗹-𝘄𝗼𝗿𝗹𝗱 𝘂𝘀𝗲 𝗰𝗮𝘀𝗲𝘀 — 𝗶𝗻 𝗺𝘆 𝘄𝗲𝗲𝗸𝗹𝘆 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿. 𝗬𝗼𝘂 𝗰𝗮𝗻 𝘀𝘂𝗯𝘀𝗰𝗿𝗶𝗯𝗲 𝗵𝗲𝗿𝗲 𝗳𝗼𝗿 𝗳𝗿𝗲𝗲: https://lnkd.in/dbf74Y9E