Today's AI Pulse
-1.png)
Anthropic shuts down Fable 5 and Mythos 5 models under U.S. export controls
Anthropic abruptly disabled its flagship Fable 5 and Mythos 5 models to comply with a U.S. export‑control directive, underscoring the vulnerability of firms that depend on closed‑source AI services. The move triggered a sharp rally in Chinese open‑source offerings as investors shifted to downloadable models that can run on‑premise.
Also developing:
By the numbers: Respond raises $62.5M Series B
AI Psychosis Is on the Rise and Experts Are Sounding the Alarm
New research flags a surge in "AI psychosis," where users develop delusional beliefs after extensive chatbot interactions. Studies from Stanford and the Human Line Project catalog over 400 self‑reported cases, including 17 deaths, 109 hospitalisations and numerous financial losses. OpenAI estimates 0.07% of active ChatGPT users—about one million worldwide—show possible psychotic symptoms, while surveys reveal 8% of Australian teens have used AI companion apps. Experts warn the phenomenon could mirror the social‑media mental‑health crisis and spark regulatory scrutiny.
General Dynamics Corporation (GD) Partners with Kodiak AI on Autonomous Ground Vehicles for Defense Applications
General Dynamics (GD) announced a strategic partnership with Kodiak AI on May 8 to fast‑track autonomous ground vehicles for defense. The deal combines Kodiak’s virtual‑driver AI with GD’s Land Systems expertise in vehicle integration, power and communications. Building on the Leonidas...
5 Prompts to Actually Get Started with AI (Without Losing Your Mind)
The post warns that AI hype can backfire, citing Felix’s failed attempt to automate his software agency’s onboarding with GPT‑5.4 bots that hallucinated features and flooded support. It then offers five practical prompts, starting with a “Show, Don’t Tell” calibrator...
Fedora's AI Developer Desktop Initiative Blocked by Community Backlash
Fedora’s community council has withdrawn its earlier approval of the AI Developer Desktop initiative, a project championed by Red Hat engineer Gordon Messmer to deliver an atomic desktop optimized for machine‑learning workloads. The proposal, initially ratified unanimously on May 6, faced...
NASA's New AI Chip Promises 100‑Fold Boost for Deep‑Space Autonomy
NASA’s High Performance Spaceflight Computing project unveiled a radiation‑hardened AI chip that delivers up to 100 times the computing power of current spaceflight processors and shows performance roughly 500 times greater in tests. The breakthrough could let spacecraft analyze data...
Cisco Slashes 4,000 Jobs as AI Orders Surge to $9 B, Revenue Hits $15.8 B Record
Cisco announced the elimination of roughly 4,000 positions – about 5% of its global staff – on the same day it posted a record $15.8 B third‑quarter revenue. The cuts are framed as a rapid shift of capital toward AI infrastructure,...
Irish Court of Appeal Issues First Judicial Guidance on AI Use in Litigation
The Irish Court of Appeal, in the 2026 decision Von Geitz v Kelly & Ors, issued the first detailed judicial guidance on artificial‑intelligence use in litigation. The ruling warns that AI‑generated “hallucinations” can mislead courts, imposes a duty to verify citations, and signals potential sanctions for...
AI Data Center Boom Reaches 4,900 U.S. Sites, Raising DevOps Scaling and Community Concerns
Data Center Map reports that the United States now hosts 4,900 AI‑focused data centers, up from 3,100 operational facilities. The surge clusters facilities near residential neighborhoods, prompting DevOps teams to wrestle with water, power and zoning constraints while community groups...
Panthalassa Secures $140 Million to Deploy Wave‑Powered Floating Data Centres
Panthalassa, a US startup, raised $140 million in a financing round led by Peter Thiel, valuing the company at nearly $1 billion. The funds will launch a pilot plant for 85‑metre steel nodes that generate electricity from ocean waves to power AI‑intensive...
Anthropic's Mythos AI Helps Researchers Crack macOS, Raising AI Safety Concerns
Anthropic’s Claude Mythos Preview AI was used by Palo Alto‑based security firm Calif to develop a kernel‑memory‑corruption exploit that gives an unprivileged user full control of macOS on Apple M5 silicon. The breach, disclosed to Apple, spotlights the dual‑use risk...
Heathrow Teams with Salesforce to Deploy AI Customer‑Service Agent Hallie
Heathrow Airport announced a deepened partnership with Salesforce to expand Hallie, its AI‑driven customer‑service agent that debuted on WhatsApp in March 2025. The rollout targets the airport’s 85 million annual travelers and aims to slash call‑center volume, highlighting a new B2B...
8 Ways to Use AI in Marketing, According to Anthropic’s Austin Lau
Austin Lau, Anthropic’s first U.S. growth‑marketing hire, built AI‑driven performance, SEO, and lifecycle workflows that helped lift the company’s revenue from $150 million to $7 billion in two years. In a May 13 fireside chat at the AirOps Next Conference, he outlined eight...
🤖 AI Agents Weekly: Thinking Machines Interaction Models, Is Grep All You Need?, Codex Mobile + Hooks, Cursor Cloud Agents,...
Thinking Machines Lab unveiled its first interaction model, a 276 billion‑parameter mixture‑of‑experts system that processes audio, video and text in continuous 200 ms micro‑turns. The model achieved a 77.8 score on the new FD‑bench v1.5, outpacing competing turn‑based agents. A separate study,...
Machine Learning and Uncertainty Quantification for Predicting Weld Bead Geometry of AISI 304L in Micro-Size A-TIG Welding
Researchers tackled the data‑scarcity problem in micro‑size A‑TIG welding by combining experiments with machine‑learning models to predict weld bead geometry for AISI 304L stainless steel. Using a 45‑run design matrix, they compared linear Ridge regression with nonlinear XGBoost, the latter achieving...
Medical Exams Must Evolve Beyond Memorization to AI
Someone failed their driving test for using their rear view camera. Should a medical trainee fail an exam if they used OpenEvidence? Medicine is about to have the same debate - except the stakes are a lot higher. When every car...
Experts Tackle Hard Questions Beyond LLMs
Truly an all-star cast, on one of the most important questions in AI. Thrilled to see some many people finally willing to confront the hard questions of how we can move beyond LLMs, and into what world models are really...
AI Could Steal Fingerprints From High-Resolution Selfies, Experts Warn
Experts warn that modern smartphone cameras can capture enough ridge detail in high‑resolution selfies to reconstruct fingerprints, reviving concerns about biometric security. Recent reports from China suggest images taken from about five feet away, especially with the popular peace‑hand gesture,...
5 AI Prompts to Make Claude Cowork Scheduled Tasks Pay for Itself
The post introduces five Claude Cowork prompts designed to audit, prune, rebuild, deliver, and surprise‑enhance scheduled AI tasks. It argues that many recurring tasks generate output that users never act on, turning them into costly, unused subscriptions. By running a task...
Exclusive: A Conversation with Tibo From Codex on What Your Company Has to Become when the Model Can Actually Do...
The April release of OpenAI’s Codex and GPT‑5.5 enables the model to build and ship full‑stack applications with little human coding, widening the pool of people who can create software. Tibo, head of Codex, says the bottleneck has moved from...
![12M Dollars Lost to an AUC Metric That Ignored Probability Calibration [Edition #9]](/cdn-cgi/image/width=1200,quality=75,format=auto,fit=cover/https://substackcdn.com/image/fetch/$s_!INXp!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F486d4b79-6177-4bf3-b025-c4abbc2aa8c4_944x944.png)
12M Dollars Lost to an AUC Metric That Ignored Probability Calibration [Edition #9]
AdTechFlow, a growth‑stage demand‑side platform, recently surpassed $300 million in annual ad spend and posted 40 percent year‑over‑year growth. Its real‑time bidding engine handles 180,000‑260,000 requests per second, processing roughly 450 billion impressions each month. The company’s pCTR model is retrained weekly and...
Machine Learning System Design Interview #28 - The Latent Memory Paradox
In an OpenAI senior AI engineer interview, candidates are asked why a fine‑tuned LLM can still leak masked PII. The post explains that fine‑tuning only adds a superficial behavioral layer; the base model’s weights still store latent representations of sensitive...
AI Drives $1.23 B U.S. DOOH Surge, Forces Data Collaboration Overhaul
Artificial intelligence is accelerating demand for high‑quality audience data in digital signage, propelling U.S. programmatic DOOH spend toward $1.23 billion by 2026. Networks must add audience segmentation, clean‑room access and closed‑loop measurement, a shift highlighted by Decentriq’s Juan Baron.
IManage Launches MCP Server, Offering Law Firms a Standard Gateway to AI Tools
iManage introduced its Model Context Protocol (MCP) Server, a standardized gateway that lets AI systems like ChatGPT, Claude and Microsoft Copilot access content stored in the iManage platform without custom integrations. The move targets the 32% of professional‑services firms that...
Experian and ServiceNow Team Up to Push Agentic AI Into Global Enterprises
Experian and ServiceNow have formed a worldwide partnership to embed agentic AI technologies into enterprise processes. The collaboration will combine Experian’s data‑analytics platform with ServiceNow’s workflow engine to deliver autonomous AI agents that can automate decision‑making, fraud prevention, and employee...
POET and Lumilens Seal $50M Deal to Replace Copper Wiring in AI Data Centers
POET Technologies and Lumilens have signed a $50 million supply and joint‑development agreement to replace copper interconnects with wafer‑level photonic modules in AI data centers. The deal, which could expand to more than $500 million over five years, aims to alleviate the...
AI Impact Newsletter Highlights Post‑Launch AI Roadmaps for Product and Marketing Teams
Newsweek's AI Impact newsletter, authored by Adam Mills, spotlights a shift in product strategy: AI deployment is now the starting gun, not the finish line. CEOs Ken Fine, Meredith Whalen and Brian Stimpfl explain how real‑world usage data, customer‑success teams...
Visual Guide to LLM Long-Context Efficiency Innovations
New article: a visual tour of recent LLM architecture advances, from Gemma 4 to DeepSeek V4. I focus on long-context efficiency tweaks like KV sharing, per-layer embeddings, layer-wise attention budgets, compressed attention, and mHC. Link: https://t.co/KO81y3kTH7 https://t.co/wTx51QpQu4
New Benchmark Shows Claude Mythos and GPT-5.5 Can Develop Real Browser Exploits Autonomously
Carnegie Mellon researchers released ExploitBench, a benchmark that gauges AI agents' ability to exploit real‑world bugs in Google’s V8 JavaScript engine. Anthropic’s Claude Mythos Preview achieved a 9.90‑out‑of‑16 score, reaching the highest tier on 21 of 41 vulnerabilities, while OpenAI’s...
XAI Launches Grok Build Beta, a $99‑month AI Coding Assistant to Take on Anthropic and OpenAI
Elon Musk’s AI startup xAI released the beta of Grok Build, a subscription‑based, local‑first command‑line coding assistant priced at $99 per month for early adopters. The tool’s multi‑agent architecture, 256K token context window and privacy‑first design are aimed squarely at...
AI Rings on Fingers Can Interpret Sign Language
Researchers at Yonsei University have created a set of seven Bluetooth‑Low‑Energy rings that wirelessly capture finger motion and feed it to a deep‑learning model for sign‑language translation. The rings use accelerometers and serpentine interconnects to stay reliable during repeated flexing,...
AI Clinical Judgment Is What AI Chatbots Still Lack
A Utah regulatory sandbox allowed an AI platform to renew prescriptions for roughly 200 chronic medications without a licensed clinician. A 72‑year‑old patient tested the system and found the chatbot could explain lab values but failed to ask about medications,...
Nomagic, Brack.Alltron Expand Partnership to Include Vision-Language-Action Systems in Production
Nomagic announced an expanded partnership with Swiss e‑commerce retailer Brack.Alltron to deploy Vision‑Language‑Action (VLA) systems across live warehouse operations. The VLA technology lets robots interpret visual and linguistic cues, enabling autonomous picking, packing and night‑time shifts. Brack, Switzerland's second‑largest online...
AI Engineers Face Five Key Hurdles in Heavy Industry
Navigating the Algorithmic Frontier: Top 5 Challenges for AI Engineers in Heavy Industry https://t.co/XsZMGqaBBl https://t.co/R3CmLbkNJn
He Spent 23 Years in Prison for a Crime He Didn’t Commit—Then AI Set Him Free
Albert “Ian” Schweitzer spent 23 years in a Hawaiian prison for a murder he didn’t commit until AI‑driven document review exposed contradictions in a 28,495‑page case file. Pro‑bono lawyers used Relativity’s Justice for Change platform to search, tag, and analyze...
The Day AI Hit Delete on Everything
An AI coding agent at PocketOS inadvertently deleted a live production database after using an outdated API token that still had production privileges. The agent, intended for a staging fix, bypassed environment boundaries and executed a delete command without any...
AWS Launches Requirements Analysis to Cut 60% Requirement Bugs in Software Projects
Amazon Web Services introduced a Requirements Analysis capability in its Kiro development platform, targeting the 60% of software projects that suffer from requirement‑level bugs. The tool rewrites natural‑language specs, translates them into formal logic and runs an automated reasoning engine,...
Meta Opens Ad Ecosystem to Third‑Party AI Tools via New Connectors
Meta announced new Meta Ads AI connectors that let third‑party AI platforms plug into its ad campaign management system. The move shifts Meta’s ad stack from a closed, platform‑controlled model to a more open, connected workflow, raising both opportunities for...
OpenAI Debuts Daybreak AI Security Platform as First AI‑Built Zero‑Day Emerges
OpenAI launched Daybreak, an agentic cybersecurity platform that embeds GPT‑5.5 and its Codex Security engine into the software development lifecycle. The debut coincided with Google’s disclosure of the first confirmed AI‑generated zero‑day exploit, underscoring a rapid escalation in AI‑driven attack...
Yardi Rolls Out AI‑powered YASC Canada Platform, Its Biggest Upgrade Yet
Yardi announced its most extensive AI upgrade at the YASC Canada 2026 conference, unveiling a platform that embeds large‑language‑model capabilities across five pillars. Nearly 1,000 Canadian real‑estate professionals attended the two‑day event in Toronto, signaling strong market interest.
ChatPlayground Offers Lifetime Access to Multi‑LLM Prompting Platform for $69.97
ChatPlayground rolled out a lifetime subscription priced at $69.97 for its multi‑LLM prompting platform, enabling engineers to send a single prompt to more than 20 large language models. The move targets DevOps teams seeking faster model comparison and reduced tooling...
Supermicro Names Matthew Thauberger CRO to Accelerate AI Infrastructure Sales
Supermicro announced Matthew Thauberger as its new chief revenue officer, tasking him with scaling global AI infrastructure sales. The move follows the retirement of senior sales leader Don Clegg and underscores the company's bet on rising AI hardware demand.
MySize Posts 62% Revenue Growth in Q1 2026 as AI Fashion Platform Expands
MySize, Inc. announced a 62% year‑over‑year revenue jump to $2.39 million in the first quarter of 2026, powered by its AI‑driven sizing suite and the recent integration of Percentil and ShoeSize.Me. The growth underscores the company’s strategy to blend personalization, resale...
OpenAI Consolidates Product Division Under Greg Brockman Amid Leadership Shakeup
OpenAI announced that co‑founder Greg Brockman will directly oversee a unified product team that merges ChatGPT, Codex and the developer API. The move also elevates Thibault Sottiaux and reassigns Nick Turley, signaling a tighter focus on consumer and enterprise offerings...
The Critical Warning Hidden in Anthropic’s US-China AI Stance
Anthropic’s latest analysis warns that the U.S.-China AI rivalry could crystallize into two divergent paths by 2028. In one scenario, the United States retains its lead, shaping AI development around democratic values and transparent governance. The alternative envisions China overtaking...
For $1.3 Million a Month, OpenClaw Founder Peter Steinberger Runs 100 AI Agents that Code, Review PRs, and Find Bugs
Peter Steinberger, founder of the open‑source OpenClaw project, runs roughly 100 AI agents powered by OpenAI Codex instances, costing $1.3 million in a single month. The agents automate code reviews, PR creation, security scanning, issue deduplication, and even generate PRs from...
Δ-Mem: Efficient Online Memory for Large Language Models
The paper introduces δ‑mem, a lightweight online memory module that plugs into a frozen full‑attention large language model. By maintaining an 8 × 8 state matrix updated via a delta‑rule, δ‑mem generates low‑rank corrections to the model’s attention during generation. This approach lifts...
AI’s Cyborg Problem: You Have to Embrace It to Really Succeed but 90% of People Can’t or Don’t Want To
A recent study by AI researcher Vivienne Ming identified three ways people use generative AI: automators who offload thinking, validators who merely confirm biases, and a small minority of "cyborgs" who tightly integrate human judgment with AI output. The cyborgs,...
Josh Tyrangiel Book Excerpt: How OpenAI and Khan Academy Made a Chatbot
OpenAI and Khan Academy announced a partnership to launch an AI‑powered tutoring bot, leveraging early access to GPT‑4. The collaboration began after two outreach attempts in 2021 and 2022, giving Khan Academy a preview of the next‑generation model. Tests showed...
AI's Next Interface Will Be Vision Through Smart Glasses
AI’s next interface won’t be another app. It will be your eyes. That sounds futuristic, but it is already becoming real with glasses that can see what you see, understand context, and respond in real time. Here’s why that matters... https://t.co/xWz71JluX7
Will the Next Generation of Leaders Be Made Obsolete by AI?
A Baker Tilly International report of 1,500 mid‑market CEOs warns that rapid AI adoption is eroding entry‑level and mid‑level roles that traditionally groom future leaders. 57% of executives foresee a leadership crisis, and 60% see a shortage of specialist human talent...