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
Gumloop Raises $50M From Benchmark to Scale AI Agents
In this episode, host Jaden Schaefer highlights Gumloop, a startup that enables every employee to become an AI‑agent builder, automating routine tasks without needing code. Founded in mid‑2023 by Max Broder Urbass, Gumloop has quickly secured major enterprise customers like Shopify and Instacart and raised a $50 million Series B led by Benchmark’s Everett Randall. The discussion emphasizes Gumloop’s model‑agnostic platform, its viral internal adoption strategy, and its positioning amid growing competition from tools such as Zapier, Dust, and native offerings from OpenAI and Anthropic.
AI’s Application Still Clouded by Hype
AI is being promoted across mining for exploration, but most tools deliver modest efficiency gains rather than breakthroughs. EY’s 2026 outlook shows 21% of miners intend to raise AI spending by over 20% despite limited returns due to siloed data...
Agents Need Vector Search More than RAG Ever Did
Qdrant, a Berlin‑based open‑source vector search firm, announced a $50 million Series B round and launched platform version 1.17. The update introduces relevance‑feedback queries, delayed fan‑out, and cluster‑wide telemetry to support the high‑throughput demands of autonomous AI agents. It highlights that agents...
Big Take: Pentagon, Anthropic and the Road to AI War (Podcast)
The Pentagon’s recent strike on Iran showcased a firepower surge that dwarfed its 2003 Iraq assault, a boost attributed in part to artificial‑intelligence integration. Hours before the attack, the Defense Department’s partnership with Anthropic ended after the AI firm raised...
Stanford Researchers Release OpenJarvis: A Local-First Framework for Building On-Device Personal AI Agents with Tools, Memory, and Learning
Stanford’s Scaling Intelligence Lab released OpenJarvis, an open‑source framework for building personal AI agents that run entirely on a user’s device. The platform introduces a five‑primitives architecture—Intelligence, Engine, Agents, Tools & Memory, and Learning—to modularize model selection, inference runtime, agent behavior, tool...
Vercel Fuels Notion’s Rise in the Agent Era
Vercel now powers the Notion agent platform. Not only are they a great company and product I use daily, I think they're going to be one of the big winners of the agent era. The main 'runtime' of an agent is English...
Spectrum Expands Partnership with RingCentral
Spectrum and RingCentral announced an expanded partnership that introduces the Unified Customer Experience (UCX) platform for Spectrum Business customers. The bundle combines RingCentral’s AI‑driven RingCX contact‑center suite and the AI Conversation Expert (ACE) with Spectrum’s managed network and onboarding services....
AI Legal Research Startup Descrybe Launches ‘Legal Reasoning’ Tool; Says It Outperforms ChatGPT, Claude, And Gemini On Bar Exam Benchmark
Descrybe introduced DescrybeLM, a purpose‑built legal‑reasoning AI that achieved a perfect 200‑out‑of‑200 score on the NCBE multiple‑choice bar exam benchmark. In contrast, leading general‑purpose models—ChatGPT 5.2, Claude Opus 4.5 and Gemini 3 Pro—missed 13‑23 questions, scoring between 88.5% and 93.5% accuracy. The study highlighted that...
AI Tracks Peer Help, Boosting Gratitude Bonuses at Google
Inside @google, we have a system for sending small bonuses to peers that helped us out. It's used often, and builds a culture of gratitude. We added an AI tool that scans your chats, emails, whatever and generates a report that...
Agentic Browsers Unnecessary; Chrome Extensions Win
ok ive seen enough evidence to call it: dedicated agentic browsers were a mistake. you can do everything u need to do as a chrome extension. Claude Cowork et al have shown the way. excited for our @felixrieseberg pod to tell the...
Better Satellite World: From Connectivity to Intelligence Episode 1: What Happens When the Network Thinks? Benny Retnamony of Quvia on...
In this inaugural episode of SSPI’s "Connectivity to Intelligence" series, host Tamara Bond‑Williams talks with Benny Rettnamone, founder and CEO of QVIA, about the challenges of handling massive data streams at the edge—on aircraft, ships, and offshore rigs—where connectivity is...
AI Bridges Contact Center Data to Automated Service Tickets
Your Contact Center is a goldmine of unstructured data, but if it isn't hitting your #ServiceManagement platform, your IT and Product teams are flying blind. AI is the bridge that turns a customer complaint into an automated internal ticket. 🔗 Build...
Marvell Partners with Mojo for Trillion‑bit Optical AI Interconnects
Marvell at the Mojo By using a grid of micro-LEDs instead of lasers, Mojo delivers 100x more bandwidth/channel + lower energy/bit. Mojo’s fiber has thousands of physically discrete channels for massively multiplexed bandwidth + redundancy for reliability in orbital data centers. Today,...
ReMix: Reinforcement Routing for Mixtures of LoRAs in LLM Finetuning
The paper introduces ReMix, a reinforcement‑learning based routing strategy for Mixture‑of‑LoRAs that eliminates the common “routing weight collapse” where a single adapter dominates. By assigning constant, equal weights to all activated adapters and training the router as a policy, ReMix...
Product & Service Launches – 3/12/2026
Morningstar, WealthBox, and FutureVault each unveiled AI‑powered tools designed to embed intelligent assistants directly into adviser workflows. Morningstar’s AI assistant integrates with its Direct Advisory Suite, enabling natural‑language navigation from client setup to portfolio analysis. WealthBox added autonomous Agents, one‑click...
Avoid Overfitting Evaluations: Blend LLMs and Human Judgment
I think there is something important going on here that is worth figuring out: 1) most models overfit to evals. that said, we need some evals to tell a model if they are good bullshit detectors before we go off and...
Naboo
Naboo.ai announced a Series B round led by Lightspeed Venture Partners, with partner Antoine Moyroud joining the deal. The AI‑driven platform automates corporate event planning across the UK and Europe, handling venues, catering, activities, and transport in a single invoice. It...
Tennessee Grandmother Jailed After AI Facial Recognition Error Links Her to Fraud
A Tennessee grandmother, Angela Lipps, spent nearly six months in jail after Fargo police mistakenly identified her via an AI facial‑recognition system as a suspect in a North Dakota bank‑fraud case. The software matched her to surveillance footage despite her never...
AI Has Created ‘Almost Infinite Demand’ for Memory Components, Dell Execs Say
Dell Technologies executives warned that artificial intelligence has generated an "almost infinite" demand for computer memory, creating a critical supply‑chain bottleneck in the United States. The shortage of DRAM and other memory chips is driving prices higher and limiting the...
Inside Brinks Home’s AI Strategy to Reduce Call Volume and Modernize Customer Support
Brinks Home leveraged generative AI to shift most support interactions from voice to digital channels, cutting call volume by over 50% while keeping satisfaction high. The company deployed a retrieval‑augmented help center, AI‑powered agent assistance, and near‑100% quality monitoring. It...
Intelligence Will Soon Be Sold Like Utilities
🚨COGNITION as a COMMODITY Here's a clip of Sam Altman describing a future in which intelligence becomes a utility that can be sold like electricity or water on a meter. To me, it was a striking comment, and I understand the...
SpaceX Space AI Ramp
SpaceX and Elon Musk’s xAI are preparing a massive orbital AI compute network, targeting a million solar‑powered satellites that act as Dyson‑swarm data centers. The plan hinges on a $50‑100 billion IPO slated for mid‑2026, which will fund new launch pads,...
We Are All AI Immigrants Now: A New Land, A Proven Mindset
Jayshree Seth argues that generative AI has turned every professional into an immigrant navigating an unmapped landscape. She introduces the GRANT framework—Grit, Resourcefulness, Adaptability, Navigation, Transcendence—as a mindset to guide AI adoption. Seth highlights that leaders must model iteration, reward experimentation,...
NATO Wants AI that Can Get in the Enemy’s Head
NATO’s Allied Command Transformation has issued an innovation challenge to develop agentic artificial‑intelligence tools that can influence adversary decision‑making on the cognitive battlefield. The solicitation, posted on Sam.gov, seeks systems that autonomously aggregate open‑source data, detect sentiment shifts, map influence...
Onyx Security Launches With $40M
Onyx Security has launched its secure AI control plane, raising $40 million from investors Conviction and Cyberstarts. The platform consolidates security, governance, and infrastructure functions to monitor AI agents across enterprises. Powered by proprietary models, it promises compliance, risk mitigation, and...
Google Leaves Door Open to Ads in Gemini
Google senior executive Nick Fox told WIRED the company is "not ruling out" ads in its Gemini AI app, marking a reversal from earlier denials. The firm is using its AI‑powered Search experience, AI Mode, as a testing ground for...
How to Analyse Customer Feedback With AI
Small and medium‑size enterprises are awash with reviews, support tickets, NPS scores and social mentions, yet they struggle to turn that volume into clear insight. Human biases—recency, negativity and confirmation—cause teams to chase loud outliers instead of representative trends. AI‑driven...
Oxa Raises $103M to Automate Industrial Driving Tasks
UK‑based Oxa raised $103 million in a Series D round led by the National Wealth Fund, with participation from NVIDIA Ventures and other investors. The capital will fast‑track development of its Oxa Driver autonomous‑driving software and the Oxa Hub cloud platform, which...
SurePath AI Advances MCP Policy Controls to Tighten the Cable on AI’s USB-C
SurePath AI unveiled MCP Policy Controls, a real‑time governance layer for Model Context Protocol (MCP) interactions. The service monitors and restricts which MCP servers and tools a codebase may use, automatically blocking or allowing payloads based on policy. In a...
Run Full LLM Offline in Any React Native App
You can now run a full LLM inside any React Native app on a 4GB phone. llama rn is a React Native binding of llama .cpp. It runs Llama, Qwen, and Mistral on-device with Metal and NPU acceleration. Fits on any modern iPhone...
Genspark.ai Extends Series B to $385M
Genspark.ai announced an extension of its Series B round to $385 million, pushing its valuation to roughly $1.6 billion. The company simultaneously launched Genspark Claw, billed as the first “AI employee,” and upgraded its AI Workspace to version 3.0, both running on a dedicated Genspark Cloud Computer....
Quit Dashboards, Master ML to Earn $200k
If you want to make $200k as a data scientist, stop making dashboards. Start doing:� • Python daily� • Data to actionable decisions� • Deploy an ML app� • Add RAG� • Add agents Need help? 👉This is how: https://learn.business-science.io/ai-register
Grok 4.20 Trails Gemini and GPT-5.4 by a Wide Margin but Sets a New Record for Not Hallucinating
XAI's latest model, Grok 4.20 Beta, posted a 48 score on the Intelligence Index, trailing Gemini 3.1 Pro Preview and GPT‑5.4, which both achieved 57. Despite lower benchmark performance, Grok 4.20 set a new non‑hallucination record, achieving a 78 % accuracy...
Gemini Pro Returns Text Instead of Images, Users Frustrated
Whenever I'm excited about something new in Gemini, I go and check it out, and it always such a sh**y experience. You can see I'm asking it to create an illustration here, and it gives me text. I'm clearly...
AI Reveals Your Brand's Weaknesses, Then Fixes Them
I ran https://t.co/f9IW600hsB on its own website. It gave itself an 𝗙. 📉 Which is hilarious, if you think about it. Even my own AI was basically like: we need to talk. 🤖 So I took its recommendations, fixed the positioning, tightened...
Robots that Learn Everyday Tasks Can Free Humans From Repetitive Work
Korea Institute of Machinery and Materials (KIMM) unveiled the RoGeTA robot task AI, a system that learns everyday activities from human demonstrations and executes them through a hierarchical task‑execution framework. The platform combines task‑extraction, virtualized training environments, and step‑by‑step reasoning,...
AI Agents Now Autonomously Execute Internet Tools in Production
An AI agent can now discover a tool on the internet, call it, and take action in production. Two years ago unthinkable, today that's a slow Tuesday night. Every enterprise will need a control plane for AI -- we backed...
Night‑time Body Pod Merges AI Hardware and Software
Exactly. We converge SW + HW since day one. The Pod runs ML on your body throughout the night. Physical AI is already here
CIQ Announces General Availability of RLC Pro AI, Enterprise Linux Built to Deliver More From Every GPU
CIQ has launched the general availability of Rocky Linux from CIQ Pro AI (RLC Pro AI), an enterprise Linux distribution optimized for AI inference and GPU workloads. The OS includes pre‑configured PyTorch, NVIDIA CUDA, and DOCA‑OFED stacks, and runs on...
Plaud NotePin: AI Wearable Boosts Meeting & Lecture Notes
I review the Plaud NotePin Wearable AI Note Taker device and find that there's a lot to like with this smart little gizmo. It's particularly helpful for meetings and lectures, as I demonstrate in this detailed video: https://t.co/Q45YtD8M9E #ai #aiWearable...
Regional AI Transformations Unveiled at MWC2026
#AI at #MWC2026: How leaders are transforming, region by region. (McKinsey) #MWC26 #MWC #JVGpost https://t.co/q6azSptnUD https://t.co/gXtGMSqk67
Unified Multimodal AI Brings Human‑like Reasoning Closer
More evidence that AI will become a unified model that can fully grasp multimodality within a single context. Similar to human thinking. Will unleash much more insightful inference, and help achieve superintelligence soon. Congrats to Google on this major advance.
AI Leverages Multi‑Industry Client Data for Insightful Answers
Prompt AI to answer questions using historic and current client data across industries and geographies. We'll analyze lessons learned and dive into AI toolsets. #AI #ClientData #Analytics https://t.co/uUyBo9DlPc
AI Gives Cities 24‑Hour Flash Flood Warnings
Protecting cities with AI-driven flash flood forecasting https://t.co/CuGhkm1sPu < I really love that we do this sort of work. @GoogleResearch is helping us give up to 24 hours advance notice for flash flood risks. https://t.co/HIVZuV6ZWP
LeCun's AI Philosophy Vs. AGI Hype Exposed
Here is an excellent paper that clearly explains the philosophy that guides @ylecun's research in AI and his new company, AMI Labs. It also perfectly expresses my complaints about the trope of artificial general intelligence--AGI, or BS for short. https://t.co/nwN07TB6Y0
Eridu May Be
Is AI networking upstart Eridu working on a high radix, waferscale switch ASIC? Maybe. Just maybe. $NVDA $AMD $AVGO $MRVL https://t.co/4337xttDy2
System Promising, but Needs 8GB M1/M2 Optimization
@Entropic_AI system looks good, but you need to figure out a path to run with 8gb M1,M2 s
Forrester Launches AI Analysis Integrated Into Enterprise Tools
It's obvious that the big analyst firms need to make a transition. Faster analysis, more original data, better sense of what's happening on the ground. @forrester is now bringing their AI-based analysis service into existing enterprise tools. https://t.co/zklowwyBTs
Seeking Latest AI-Generated Music Hits and Tools
What's the SotA for AI music generation these days? Any AI generated bangers you've listened to lately?
Separate Test Centres Streamline AI Assessment, Reduce Faculty Load
Interesting webinar on assessment in AI, recommending separate TEST CENTRES. Like driving, get expert tuition, then sit test separately. Removes summative assessment from faculty.