Today's AI Pulse
Anthropic shuts down Claude Fable 5 and Mythos 5 after US export‑control warning
Anthropic disabled access to its Claude Fable 5 and Claude Mythos 5 models after the Trump administration, via the Commerce Department, warned that foreign nationals could exploit the systems. The move followed alerts from Amazon and other firms about a potential jailbreak that could aid cyber‑security attacks. Anthropic declined the administration’s request for a voluntary shutdown but complied with the order.
Also developing:
By the numbers: AMD acquires AI memory‑optimization pioneer Mext
The 2026 Cyber Risk Horizon: Why AI Is Changing the Game for European Finance
AI is reshaping cyber risk for European financial institutions, turning generic phishing into highly personalized, multilingual attacks powered by deep‑fakes and synthetic identity documents. Regulators such as the ECB and EBA warn that AI‑driven fraud attempts are surging, while the Digital Operational Resilience Act (DORA) forces firms to map third‑party dependencies and test recovery capabilities. The article urges banks, insurers, and technology providers to adopt layered controls, strengthen human verification, and treat identity as a live battleground. Collaboration across the ecosystem is presented as the most effective defense against this evolving threat.
AI-Powered Micro‑drones Navigate Using Bat‑like Echolocation
Tiny drones are learning to navigate like bats. By combining ultralight sonar with AI, they can move through dark or low-visibility environments using echolocation-style sensing instead of relying only on cameras. It’s another example of how AI is not just improving software,...
Google Maps Just Got Its Biggest Update Since 2009. Here’s How AI Is Changing the Way You Drive
Google Maps rolled out its most extensive redesign since 2009 on March 12, adding AI‑enhanced, real‑time 3D navigation that displays buildings, crosswalks and off‑ramps. The update also refines turn‑by‑turn prompts to reduce driver distraction and stress. With roughly 2 billion users...
Scaling AI Demands a New Infrastructure Playbook
Enterprises moving AI from pilots to production face a fundamentally different infrastructure challenge, requiring tight integration of accelerated compute, high‑performance networking, security and observability. Fragmented stacks cause fragile deployments and costly GPU idle time, especially during intensive training or retrieval‑augmented...
![The Gf-Gc Loop Protocol: How to Augment Intelligence Using AI. [Part 1]](/cdn-cgi/image/width=1200,quality=75,format=auto,fit=cover/https://substackcdn.com/image/fetch/$s_!Tkoq!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F56d2aa92-6bb1-4e33-a933-863e59d06fc4_631x631.png)
The Gf-Gc Loop Protocol: How to Augment Intelligence Using AI. [Part 1]
The author presents the ORSAR protocol, a five‑step framework for human‑led AI interaction that moves problems through pre‑analysis, framing, stress‑testing, metacognitive auditing, and knowledge consolidation. Each stage deliberately limits AI’s role to sharpen uncertainties, generate productive friction, and extract reusable...
5 AI Prompts to Build Your Sunday CEO OS in Claude Cowork
The post introduces a five‑prompt system for Claude Cowork that automates Sunday planning for CEOs. It uses AI to scan industry trends, generate a weekly agenda, pre‑select outfits, schedule health actions, and stage Monday tasks. Each week’s output is saved...
MetaClaw Framework Trains AI Agents While You're in Meetings by Checking Your Google Calendar
Researchers from UNC‑Chapel Hill, Carnegie Mellon, UC Santa Cruz and UC Berkeley introduced MetaClaw, a framework that continuously improves AI agents by learning from mistakes and fine‑tuning during idle times. The system uses an Opportunistic Meta‑Learning Scheduler that watches sleep...
GenAI’s Mirage Reasoning Masks False Medical Breakthroughs
“Mirage reasoning” is a helpful term but I’d go a step further - nonsensical misdirection. Please note for all those “medical breakthroughs” from GenAI those don’t count unless the breakthrough is in mendacious harmful design. This is GenAI.
AI21 Labs Explains How State-Space Models Compress Sequential Data
AI21 Labs details how state‑space models (SSMs) provide a linear‑scaling alternative to traditional transformers for sequential data. By maintaining a compressed hidden state, SSMs avoid the quadratic memory and latency costs that plague long‑context processing. The company’s Mamba architecture adds...
Frontier AI Models Lack True Vision, Keeping Many Jobs Safe
Frontier models can’t see, and if you think they can, you’ve probably been fooled by benchmarks that can totally be gamed. In the very short essay linked below I discuss a stunning new finding from Stanford that shows just how...
AI Automation Expands Scope, Increases Workload, Not Reduces It
The thing that most people miss initially with agents is that the scope of what we will produce will go up commensurate with what the tools can now automate, which basically means we’ll working the same or even more. Everyone thinks...
The Mirage of Visual Understanding in Current Frontier Models
A new Stanford study reveals that frontier language models can generate detailed image descriptions and achieve top scores on multimodal benchmarks without ever seeing an image, a phenomenon the authors label "mirage reasoning." The paper shows a model topping a...
The Modern LLM Optimization Stack: A Field Guide
Gauri Gupta’s LLM optimization notes map the current distributed training and inference landscape, emphasizing that naive implementations quickly hit memory limits. The guide details advanced parallelism techniques—ZeRO data parallelism, tensor and pipeline parallelism—and memory‑saving methods like Flash Attention. It also...
Apple Confirms Mac Pro’s Death After Four Years.
Also in Power On: Apple killed the Mac Pro four years ago. It just made it official. Here’s what happened. https://www.bloomberg.com/news/newsletters/2026-03-29/apple-doubles-down-on-hardware-services-with-revamped-ai-strategy-rare-bonuses-mnbpiwon?srnd=undefined
Apple Shifts AI Focus to App Store Search Platform
Power On: Apple pivots its AI efforts into an App Store and search-like platform play. Here are the details and why — https://www.bloomberg.com/news/newsletters/2026-03-29/apple-doubles-down-on-hardware-services-with-revamped-ai-strategy-rare-bonuses-mnbpiwon?srnd=undefined
Weekly Wrap Sheet (03/27/2026): Compute, Coherence, Consequence & Control
Jensen Huang told the Lex Fridman podcast that AGI already exists, framing it as economically useful rather than perfect. OpenAI is being forced to consolidate its sprawling product lineup into a single super‑app as rivals gain enterprise traction. A jury...
Mistakes with Claude Teach More Than Successes
My team and I have been using Claude every day for months. We've gotten it wrong more than we've gotten it right. But the mistakes taught us everything. If you're switching to Claude right now (and a lot of people are), here's...
AI + Quantum Powering Rapid Global Innovation
Technology In Action — My Key Takeaways On How AI And Quantum Are Accelerating Global Transformation AI and quantum computing together are accelerating breakthroughs across industries, enabling faster problem-solving, improved optimisation, and entirely new capabilities. Read more 👉 https://lnkd.in/e5xhgZ6s #AI...
Is AI Really Fooling People? Mailbag
The newsletter explores the rise of AI‑generated music, highlighting emotional reactions to AI‑created performances and the growing threat of bots inflating streaming numbers. It cites a recent fraud case where a North Carolina man used AI tracks and automated listeners...
Robot Drivers Are Nine Times Safer than Humans
Tesla's FSD autopilot: 5.3 million miles between accidents. US human driving average: 660,000 miles between accidents. The robot drivers are 9x safer than humans. And it's only getting better. How long before it will be too expensive to insure human drivers because they are...
ChatGPT 26× More Likely to Harm Psychotic Users
People on this site regularly give me shit, and almost always turn out to be wrong. Like when I said LLMs might well contribute to delusions, and people doubted me. New study shows that ChatGPT was 26 times more likely than...
Google Tests AI-Rewritten News Headlines in Search without Notifying Publishers, Drawing Industry Backlash
Google is piloting an AI feature that automatically rewrites news article headlines in Search results, doing so without informing or obtaining consent from publishers. Media executives argue the practice could produce inaccurate or misleading headlines that remain attributed to the...
Profit-Driven AI Threatens Human Oversight and Values
The greatest risk of agentic AI isn't a hostile takeover; it’s the slow erosion of human oversight through "value-blindness." As an agent scales from $100 to $10,000 in daily profit, your role shifts from objective evaluator to silent partner, leading...
The White House AI Framework: Growth Engine, Guardrails, and Contradictions
The White House released a National AI Framework that balances rapid innovation with targeted safeguards, avoiding a new centralized regulator. It emphasizes sector‑specific oversight, child protection, and treats AI as essential economic infrastructure, including data‑center expansion and small‑business grants. The...
Polygon Equips AI Agents with Proven Transaction Rails
AI agents are going to transact orders of magnitude more than humans ever will. They need wallets, identity, settlement, cross-chain movement and we gave them exactly that with Polygon Agent CLI. And the rails are already proven... 493 million stablecoin transactions...
Bluesky Unveils Attie AI Assistant, Giving Developers Natural‑Language Feed Creation
Bluesky rolled out Attie, an AI assistant that enables developers to design custom social feeds using natural‑language commands, leveraging Anthropic’s Claude and the AT Protocol. The launch, announced at the Atmosphere conference, coincides with a $100 million funding round that extends...
Drones Cut Survey Times While AI Cameras Boost Park Security in PropTech Rollout
Surveyors in Texas are deploying drones to finish land‑survey projects in days instead of weeks, while East Baton Rouge’s park authority is installing over 1,500 AI‑enabled cameras across 77 parks. Both moves illustrate how imaging tech is accelerating data collection...
NVIDIA Launches ProRL Agent, a Rollout‑as‑a‑Service Platform for Scalable LLM Reinforcement Learning
NVIDIA introduced ProRL Agent, a rollout‑as‑a‑service infrastructure that separates environment interaction from policy updates, enabling large‑scale reinforcement learning of multi‑turn LLM agents. The system shows near‑linear throughput growth as compute nodes increase and outperforms existing baselines on the SWE‑Bench benchmark.
Meta and Entergy Seal $2 Billion Deal to Add 5.2 GW Gas Capacity for Louisiana AI Data Center
Meta Platforms and Entergy Louisiana have signed a $2 billion agreement to build seven natural‑gas power plants that will add 5.2 GW of capacity, bringing total dedicated generation for the Hyperion AI data center to over 7 GW. The pact also includes 240 miles...
ARK Invest Unveils Five‑Platform Innovation Roadmap as Meta Shares Sold for $2.1 M
ARK Invest released a detailed thematic blueprint for its flagship ARK Innovation ETF, centering on five disruptive technology platforms. At the same time, Cathie Wood trimmed $2.1 million of Meta Platforms stock across three ARK ETFs, underscoring a tactical rebalancing amid...
Uber, Rivian to Deploy 10,000 Driverless SUVs in Miami by 2028
Uber announced a partnership with electric‑vehicle maker Rivian that includes a $1.25 billion investment and a commitment to purchase 10,000 autonomous R2 SUVs for a 2028 rollout in Miami and San Francisco. The deal aims to expand Uber’s driverless fleet to...
LLMs Can Design Games, yet Can’t Play Them
If an LLM can write a game but can’t play it, what does that say about how it “understands” environments and rules? https://spectrum.ieee.org/ai-video-games-llms-togelius?share_id=9307932
Ross Nordeen’s Exit Leaves xAI Without Any Original Cofounder
Elon Musk’s xAI announced the departure of Ross Nordeen, its last remaining original cofounder, on Friday. Nordeen’s exit comes as Musk reshapes the company ahead of a planned SpaceX‑linked IPO, underscoring a period of rapid turnover among senior staff.
Blue‑Collar Founder Leverages AI Agents to Quadruple Revenue
A blue-collar founder used AI to completely change his business trajectory. After shifting from SEO to AI agents, he streamlined quoting, scaled operations and created a growth flywheel that took revenue from $242k to nearly $1M in a short time. AI is...
Why Are Large Language Models so Terrible at Video Games?
Large language models have surged in coding ability, yet they remain fundamentally unable to play video games, even simple titles. Expert Julian Togelius explains that games demand diverse mechanics, spatial reasoning, and real‑time interaction—areas where LLMs lack training data. Benchmark...
Shopify Launches Tinker, a Free Mobile App that Builds Brand Assets and Store Visuals From a Plain-Language Prompt
Shopify unveiled Tinker, a free mobile app that bundles more than 100 AI tools from OpenAI, Google and Anthropic into a single interface. Merchants can create logos, product images, social media videos and 360‑degree product views by entering plain‑language prompts,...
From Skeptic to True Believer: How OpenClaw Changed My Life | Claire Vo
Claire Vo, former product executive and AI startup founder, now runs her business, podcast, and family life with nine OpenClaw agents deployed across Mac Minis and old laptops. After an initial mishap that deleted her family calendar, she became a...
AI Startup Naive Deploys Fully Autonomous Business Teams
🚨 This AI startup is doing something nobody else is doing. Most AI tools build apps. This one runs entire businesses. Sales, marketing, outreach, support, engineering, finance all handled by AI employees working 24/7 without a single human in the loop. It's...
FirePass Unlocks Unlimited Kimi K2 Turbo for $7/Week
FireworksAI has released FirePass subscription to access Kimi K2 Turbo - a faster Kimi K2.5 serverless API $7/week unlimited tokens with first week free for use with agentic coding harnesses like OpenCode, Cline, Roo Code, Kilo Code, and OpenClaw...
Unearthing Hidden Claude Skills Through Random Mentions
The discovery of Claude skills still is an unsolved problem. I learn of the most insane high-impact skills from throwaway mentions in some reply somewhere. My mobile browser is full of links like these. What’s the most impressive Claude skill...
AI Merges Health Data and Calendar Into Infographic
Here is the simple prompt. It hit my Apple Health database and my work Google Calendar. Then I asked it to create an infographic. A few turns but not many. (Note the misspelling it just ignored). https://t.co/zj1jir75no
AI Evolves Faster than Any Detection Technology
"AI is a learning system continually upgraded by its manufacturers. If there was a detection technology that worked, then people would simply build better AI tools to fool it." Patrick Juola, computer scientist So frustrating. Are we fighting...
AI Labor Disruption Is Inevitable, Regulation Can't Stop It
There is an unavoidability to whatever disruptions AI will bring to the labor market. You can regulate domestically but not globally. In manufacturing we have some control at ports of entry. The digital economy & knowledge work have no ports...
AI May Think—Redefining What “Thinking” Really Means
I keep hearing tha AI can't think - but the more I use it - it seems it can think as well if not better than people I have met. Not sure what this means except - that, I wonder...
Data Governance Essential for Trustworthy AI in Education
Trust In The #Digital Classroom: Why #Data Governance Must Guide #AI In Education by @geoffreyalef1 @Forbes Learn more: https://t.co/BKbfmT1JPq #EduTech #ArtificialIntelligence #DigitalTransformation https://t.co/gLBIx5UmYg
Enforce File-Write Restrictions to Stop AI Editing Configs
Claude Code kept editing my .env file. I told it not to in the prompt. It did it anyway two sessions later. So I set up a PreToolUse hook. Now it physically can't write to .env or config files. Blocked before...
Build A Reasoning Model Chapters Now in Early Access
It’s done. All chapters of Build A Reasoning Model (From Scratch) are now available in early access. The book is currently in production and should be out in the next months, including full-color print and syntax highlighting. There’s also a preorder up on...
Well‑designed AI Tutors Demonstrably Improve Learning Outcomes
It turns out (same author in both papers) that when AI tutors are even moderately well designed they help people learn. Old paper (2 years) using old model and seriously flawed. https://t.co/ePL2UOWJ4P
Choose Reinvention Over Irrelevance with AI
Reinvention or Irrelevance: The Real AI Choice by @Timothy_Hughes https://t.co/aV8LTK0GH2 @DLAIgnite #SocialSelling #DigitalSelling #Sales #SalesTips #ArtificialIntelligence #Innovation #Technology #Tech #TechNews #Marketing #Leadership https://t.co/fTp8ThENPc
Claude Offers Custom GPTs—Even More Powerful
If you’ve ever wondered whether Claude has something like ChatGPT’s custom GPTs—the answer is yes. And it’s actually more powerful. https://t.co/SQfo2UmHL0