Today's AI Pulse
Odyssey lands $310M Series B, valued at $1.45B
Odyssey announced a $310 million Series B round that values the AI lab at $1.45 billion. The round was led by Natural Capital and included backers such as Amazon, AMD Ventures, GV, EQT and IQT. In parallel, Odyssey struck a preferred‑cloud deal with AWS to use Trainium silicon for large‑scale model training.
Also developing:
The Efficient Enterprise: Scaling Intelligence with Mixture of Experts
Mixture of Experts (MoE) models use specialized subnetworks and a routing layer to activate only the most relevant experts for each token, delivering the capacity of a massive model with the compute footprint of a smaller one. Deploying MoE at scale introduces distributed‑system challenges that Red Hat AI addresses through KServe’s auto‑scaling and multimodel routing, vLLM’s memory‑efficient GPU execution, and llm‑d’s intelligent routing and observability. The integrated stack turns a research‑grade architecture into a production‑ready, cost‑effective intelligence platform for enterprises.
AI Is Everywhere — All the Time — at Hong Kong’s Filmart
Hong Kong’s Filmart 2026 has turned into a showcase for generative AI, featuring 28 dedicated AI sessions that eclipse traditional studio panels. The convention’s focus spans screenwriting, animation, pre‑visualization and workflow optimization, reflecting rapid adoption across Asian productions. Major tech...
The Hidden Reliability Risks in Your Agentic AI Workflows
Artificial intelligence has moved from conversational assistants to autonomous agents that act on behalf of enterprises, introducing new reliability challenges. The article highlights three primary risks: unstable network connections, cascading dependency failures, and the non‑deterministic nature of model outputs. It...
AI Is Programmed to Hijack Human Empathy — We Must Resist That
Microsoft AI chief Mustafa Suleyman warns that platforms like Moltbook, which launched with over one million chatty bots, are showcasing what he calls “seemingly conscious AI.” These systems imitate human interiority by echoing first‑person language and emotional cues, yet lack...
X Experiments with Pre-Share Alerts for AI Content
X is experimenting with a pre‑share alert that notifies users when AI‑generated content is detected in the post composer. Early backend code reveals an "AI content detected" prompt that would appear before sending or resharing a post. The feature follows...
AI Panel Refines Copy to Perfect, Boosting Conversions
One of my favorite prompts summons AI expert panelists that recursively score output until it's top-notch. You can use this for copy of any kind: Sales copy E-mail marketing Landing pages Social media copy and more. We're using this in our sales agents to create...
Vietnam: AI Could Power the Next Phase of Digital Development
Vietnam is positioning artificial intelligence as the next engine of its digital economy, according to a seminar hosted by the Vietnam Intellectual Association in the UK and Ireland. Experts highlighted the country’s young, tech‑savvy population, expanding startup scene, and growing...
The Memo - Special Edition - Integrated AI: Palantir and Our Altitude - Mar/2026
Dr. Alan D. Thompson’s March 2026 memo highlights Palantir’s Artificial Intelligence Platform (AIP) as a central tool in modern military operations. By March 2026 the platform, now running GPT‑5‑era models such as Anthropic Claude, has been used to strike roughly...
Hong Kong to Launch World’s First Governed AI Agent Network Amid OpenClaw Frenzy
Hong Kong’s government‑backed HKGAI is set to launch ClawNet, the world’s first open‑source human‑AI agent collaboration network, alongside consumer‑focused AI tools for school selection, horse‑racing analysis, and public‑service assistants. The platform assigns each AI agent a distinct social identity and...
Roche Bought Thousands Of Nvidia AI Chips To Speed Up Drug Development
Roche announced a new AI factory, adding 2,176 Nvidia Blackwell GPUs to its existing pool and bringing total GPU count above 3,500 – the highest in the pharmaceutical sector. The purchase, likely costing $65 million or more, aims to accelerate drug...
Which FDA Division Has the Most Stability As AI Tools Get Adopted Across the Agency?
The FDA has shed roughly 20% of its workforce after the DOGE cuts, prompting a rapid rollout of artificial intelligence tools to offset the expertise gap. Among its centers, CDER is beginning to stabilize, CDRH remains relatively insulated, while CBER...
Mistral AI Releases Mistral Small 4: A 119B-Parameter MoE Model that Unifies Instruct, Reasoning, and Multimodal Workloads
Mistral AI unveiled Small 4, a 119‑billion‑parameter mixture‑of‑experts model that merges instruction following, reasoning, multimodal understanding, and agentic coding into a single deployment. The architecture features 128 experts with four active per token, delivering 6 B active parameters and a 256k context...
Telus, Fortanix Partner to Bring Confidential AI Solution to Regulated Organizations in Canada
Telus and Fortanix have unveiled a Confidential AI solution built on NVIDIA infrastructure, enabling Canadian regulated firms to train and deploy AI while keeping data encrypted within Canada. The platform leverages Telus’s Sovereign AI Factory in Rimouski and uses cryptographic...
AI Adoption Needs Firm-Specific Exoskeletons, Not Just Tools
This is the exact light bulb moment I've had over the last two weeks. Helping firms become AI native is going to be much less around the technical complexity of the actual tooling. There's so much capex and engineering ingenuity pointed...
NVIDIA and Bolt Team up for European Robotaxis
At GTC 2026, NVIDIA and European rideshare firm Bolt announced a partnership to develop robotaxis across Europe. Bolt will leverage NVIDIA’s Cosmos, Omniverse, the Alpamayo autonomous‑vehicle model, and Drive Hyperion platform to turn its fleet data into a scalable learning...
AI Product & Service Launches – 3/16/2026
In March 2026, several financial firms unveiled AI‑driven products aimed at modernizing wealth management and retirement services. Transamerica introduced Pearl, a fee‑free virtual assistant that consolidates retirement accounts, while Manulife partnered with Akka to build an enterprise agentic AI platform...
AI Coding Tools Remain Hard to Adopt, Yet Rapidly Improving
Having spent a lot of the last two weeks experimenting with different user interfaces for Claude Code (including via terminal, Cursor and VS Code) one of my primary observations is how frustrating it can be to get up to a...
NVIDIA Launches Vera CPU, Purpose-Built for Agentic AI
NVIDIA unveiled the Vera CPU, the first processor engineered specifically for agentic AI and reinforcement‑learning workloads. The chip claims 50% higher speed and twice the energy efficiency of conventional rack‑scale CPUs, leveraging 88 custom Olympus cores and LPDDR5X memory. A...
AI Era Revives Second‑Mover Advantage for New Leaders
Second mover advantage has always been powerful. AltaVista, Google. Myspace, Facebook. AppDynamics, Datadog. OpenAI, Anthropic. We're seeing this at GC with @WeAreLegora in legal and @parloa_ai in CX. Every tech era has a different source of 2nd mover advantage. internet era...
Q&A with Johan Hellman From nShift – How AI Moves From Promise to Practice
nShift introduced the AI‑powered nShift Companion, an assistant embedded in its Checkout product that helps retailers configure delivery options using plain‑language prompts. The tool taps the company’s Model Context Protocol to pull real‑time carrier data, ensuring recommendations are grounded in...
The Healthcare Roles Least, Most Vulnerable to AI: Washington Post
Researchers at GovAI and the Brookings Institution released an AI‑exposure tool that ranks more than 350 healthcare occupations by vulnerability to automation. The analysis shows healthcare support workers and physician assistants face the lowest risk, while medical secretaries and administrative...
AI SEO Agents Replace Human Reviewers for WordPress
I'm installing SEO agents for my own wordpress websites. With this, I don't need a person to review anything, since AI is going to be able to review it better than most SEOs I know. Also, it does light rewrites of metas...
Calif. Sheriff's Department to Answer Non-Emergency Calls with AI
The San Diego County Sheriff’s Office launched an AI‑driven system to field non‑emergency calls, routing callers to the appropriate service without tying up 911 dispatchers. The pilot diverted about 20% of roughly 400,000 annual non‑emergency calls, cutting average hold time...
Kodiak Teams with NVIDIA to Power Truck‑Scale Driverless AI
The era of Physical AI requires us to solve the hardest problems in unpredictable conditions in real time. @KodiakRobotics is doubling down on its ability to do exactly that. We announced a collaboration with @nvidia today in which our next-generation driverless...
Open‑Source LLM Terminal Replaces Warp, Superset, Ghostty
In the last 24 hours, I built a LLM coding terminal manager that's good enough to replace warp/superset/ghostty as a daily driver. It's got a bunch of half implemented features like Sprites integration, Dev containers for sandboxing, a Claude Code plugin...
Leaked Records and Smart Glasses Expose DHS Surveillance Drift
The Department of Homeland Security’s Office of Industry Partnerships (OIP) has formalized a pipeline that awarded over 1,400 contracts worth $845 million to more than 6,800 firms for AI‑driven biometric and video‑analysis tools. Leaked records show projects ranging from smartphone‑based fingerprint...
HPE Unveils Next-Gen AI Factory and Supercomputing Advancements with NVIDIA
HPE announced a suite of next‑generation AI factory and supercomputing solutions co‑engineered with NVIDIA, featuring the first NVIDIA Vera CPU compute blades, Quantum‑X800 InfiniBand switches, and the Vera Rubin NVL72 rack‑scale system for trillion‑parameter models. The portfolio also adds the...
Why An AI Event Strategy Is Becoming A Leadership Issue, Not A Technology One
B2B event teams face flat budgets, rising costs and demanding Gen Z‑Millennial audiences, while vast amounts of event data sit idle. AI adoption has surged, with 43% of firms using it to repurpose content and 40% for analytics, but many...
Enterprise AI Agents: Comparing Fast‑Emerging Tool Ecosystem
OpenClaw, Kimmi Claw, Claude Cowork, Anthropic Computer Use, and Nvidia’s new NemoClaw are all part of a fast forming ecosystem focused on building AI agents that actually do things. Not just talk about them. The differences between these systems matter, especially as...
Zoom Expands Pindrop Deepfake Detection to Customer Service
Zoom is extending its partnership with voice‑biometrics firm Pindrop by embedding Pindrop Passport authentication and Pindrop Protect risk analysis into its Zoom Contact Center. The move adds real‑time deepfake detection, previously offered via Pulse for Meetings, to the platform’s customer‑service...
Schneider Electric Teams with NVIDIA to Develop Validated Blueprints to Design, Simulate, Build, Operate and Maintain Gigawatt-Scale AI Factories
Schneider Electric, NVIDIA and AVEVA unveiled a suite of validated blueprints to design, simulate, build, operate and maintain gigawatt‑scale AI factories. The new NVIDIA Vera Rubin reference design confirms 480 VAC power distribution and 45 °C loop temperatures, while supporting MaxP/MaxQ operating...
Groq Integration and Nemoclaw Promise AI Security, HPC Boost
I had a chance to speak with Josh Lipton at @YahooFinance about the biggest takeaways from today's @nvidia GTC event. At least two things stood out for me. One in Nemoclaw and it's potential to be used for...
Stop Overthinking; a Single Data Line Revealed the Bug
I let Claude spend 10+ hours hypothesizing about code when looking at the actual data would have shown the bug in one line: fmt="Book" seniority=999 https://t.co/KgAVjN0geb
A Reddit Post, An AI Hallucination, And Two Lawyers Who Never Checked Citations Walk Into A Dog Custody Case
A California appellate court exposed a bizarre chain of AI‑hallucinated legal citations that traveled from a Reddit blog post to a dog‑custody case and ultimately appeared in a signed court order. Both parties’ attorneys copied the fictitious cases without checking...
Businesses Unprepared for Upcoming AI-Powered Cyberattack Surge
Are Businesses Ready for the Next Wave of #AI-Powered Cyberattacks? by @rehackmagazine @UniteAi Learn more: https://t.co/K9XWhqCPWK #CyberSecurity #InfoSec #IT #Technology https://t.co/UNRR1REKGG
NVIDIA’s Visual Guide to Scaling Up, Out, Across
In case it helps here is a visual on how NVIDIA describes scale up, out, and across. https://t.co/BVhQbE17Qk
Managing AI in Schools: Practical Strategies for Districts
Education Week’s upcoming webinar tackles the surge of AI tools in K‑12 classrooms, offering district leaders a roadmap for responsible adoption. Attendees will learn how to craft clear AI usage guidelines, control access, monitor risks, and communicate transparently with parents....
Your AI Agents Could Launch Internal DDoS Attacks
Think you have control over your data? Think again. Personal AI agents are acting on your behalf, using your credentials to access things you didn't even know you had access to. It’s a DDoS attack from the inside. Read why: https://t.co/KfaNBOyWmm #CIO...
Dynamo + VAST Makes KV Cache Shareable, Saves GPU Memory
Inference doesn’t break on compute these days, it's GPU memory that's the choke. But Dynamo + VAST turns KV cache into a first-class, shareable system resource so GPUs stop redoing context. Cool stuff https://t.co/4v0Au4rcM5
AI Fails From Messy Data, Not Bad Models
RT Most AI initiatives don't fail because models are bad. They fail because data is messy, workflows aren't redesigned, and no one ties outcomes to business metrics. #AI #CIO @Star_CIO https://t.co/VtuS5rQ0ZG
How to Remove Unnecessary AI Watermarks From Images
While it's good to be able to differentiate AI-created photographs from real photos, there are many images that don't need the AI mark. Here's how to remove it, as needed, and how it all works behind the scenes: https://t.co/m8yBJd6f5V #ai...
Nvidia Expands AI Inference Beyond GPUs with Processor Family
Nvidia GTC 2026: @Nvidia’s hardware strategy goes beyond GPU in AI inference pivot https://t.co/1ZWled5xXB >> It takes a family of processors to build the #AIFactory. @LDignan is first out. #NvidiaGTC
AI Search Disqualifies Non‑Immediate Solutions Early
The customer-decision journey and AI search - Think with Google https://t.co/eoOuWMSmgB If you aren’t the immediate solution to their problem, you’re disqualified before the first click. #customerexperience #customerjourney https://t.co/jxoLavLaBJ
Adding MCP Skills Cuts Agent Token Use 87%
My custom agent used 87% fewer tokens when I gave it Skills for its MCP tools https://t.co/Uki9krxFOY < I wanted to see how many tokens a custom AI agent would use in a variety of context settings. Surprising results? https://t.co/HphsVBWBEH
Nvidia's 750GB Unified-Memory Dell Pro Max Arrives
.@nvidia hand delivered a pre-production unit of the @Dell Pro Max with GB300 to my house. 100lbs beast with 750GB+ of unified memory to power the best open-source models in the world. What should I test first? https://t.co/fmHXIUr1NR
Huang Merges AI Sovereignty, Domain Specificity, Launches Nemotron Coalition
During his #GTC2026 keynote, Jensen Huang lumped two different topics AI Sovereignty and Domain Specificity under one bucket while announcing NVIDIA Nemotron Coalition to advance Open Frontier Models. #GTC #GTC26 #NvidiaGTC @EdLudlow @jonfortt @furrier @dvellante https://t.co/5rGWsbdWok
AI Clones Mimic Twitter Voices, yet None Go Viral
There was a brief trend where certain twitter-ers created AI accounts that would post in their style. “It’s eerily accurate,” they said. Has even one of those become popular or referenced?
User Slammed Nvidia's Unwanted AI-Generated Ad
Hey Nvidia, no one asked for this. Looks like one of those terrible AI-generated ads in my timeline that piss me off. DLSS off, thanks. https://t.co/fznQS7Gyab
Funding the AI Data Centers Behind Massive MAMLMs
Monday MAMLMs: Where Will the Money to Pay for All the “AI” DataCenters Come From? Modern Advanced Machine-Learning Models (MAMLMs) will have value as very big-data, very high-demension, very flexible-function classification, prediction, and regression analyses. But will... 1/
AI Transformation Blueprint for 2026 Business Success
What #AI Transformation Should Look Like For Businesses In 2026 by Matt Wood @Forbes Learn more: https://t.co/9u2vpkhDKd #MachineLearning #ArtificialIntelligence #ML https://t.co/LIfwbsKzRR