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:
Why AI Is Both a Curse and a Blessing to Open-Source Software - According to Developers
AI is reshaping open‑source development, delivering rapid security bug discovery for projects like Firefox while also flooding smaller maintainers with low‑quality, AI‑generated reports. Anthropic’s Claude helped Mozilla identify high‑severity bugs in weeks, yet cURL’s security team faces a deluge of false positives, likened to a DDoS of bogus reports. Linux maintainers are integrating AI into patch triage and back‑port workflows, viewing it as a productivity tool rather than a code‑authoring replacement. The divergent outcomes underscore that AI’s impact hinges on disciplined use, human oversight, and clear accountability.
Microsoft’s Copilot CoWork Runs on Anthropic’s Claude AI
𝗖𝗹𝗮𝘂𝗱𝗲 𝗔𝗜 𝗶𝘀 𝗾𝘂𝗶𝗲𝘁𝗹𝘆 𝗺𝗼𝘃𝗶𝗻𝗴 𝗶𝗻𝘁𝗼 𝘁𝗵𝗲 𝗲𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲. This week, 𝗠𝗶𝗰𝗿𝗼𝘀𝗼𝗳𝘁 introduced 𝗖𝗼𝗽𝗶𝗹𝗼𝘁 𝗖𝗼𝘄𝗼𝗿𝗸, a new 𝗮𝘂𝘁𝗼𝗻𝗼𝗺𝗼𝘂𝘀 𝗔𝗜 𝗮𝗴𝗲𝗻𝘁 designed to work across the entire 𝗠𝗶𝗰𝗿𝗼𝘀𝗼𝗳𝘁 𝟯𝟲𝟱 environment. The system can 𝗽𝗹𝗮𝗻 𝗮𝗻𝗱 𝗲𝘅𝗲𝗰𝘂𝘁𝗲 𝗺𝘂𝗹𝘁𝗶 𝘀𝘁𝗲𝗽 𝗼𝗳𝗳𝗶𝗰𝗲 𝘄𝗼𝗿𝗸𝗳𝗹𝗼𝘄𝘀 across Outlook, Excel,...
Weekly AI and Health Tech Trends for Clinicians
Every week, I track the signals that show where medicine is really going from the latest AI breakthroughs and digital health tools to the cultural shifts redefining patient care. The Medical Futurist Newsletter brings you: ✅ Curated insights on the most impactful...
LLM Agents Interview Questions #16 - The Vision Encoder Scaling Trap
In a mock Google DeepMind interview, candidates are asked why upgrading a geometry auto‑formalization pipeline from a 70B text‑only LLM to a state‑of‑the‑art vision‑language model (VLM) only yields a 20% success rate. Most answer that the vision encoder loses spatial...
FCC Chair Calls on AI Regulators to Take Humble Approach
FCC Chair Brendan Carr urged regulators to adopt a humble, light‑touch approach to AI policy, likening it to the United States’ hands‑off stance on the early internet. He warned that over‑regulation could stifle innovation and emphasized AI’s massive upside. Carr...
Polymarket Enlists Palantir, TWG AI to Monitor Sports Bets – Bloomberg
Polymarket is partnering with Palantir Technologies and TWG AI to monitor its sports‑betting contracts for suspicious activity. The firms will screen participants against banned‑list databases and flag potential insider‑trading violations. This monitoring system will be deployed on a new U.S.-regulated...
Writing as Thinking
The article challenges the notion that blogging is a waste of time, especially as AI tools like WordPress’s new "AI Tool" promise instant content creation. It argues that writing serves as a thinking process, helping authors surface incoherent ideas and...
Podstock Launches ‘Agent’, an AI Assistant for Podcast Analytics & Operations
Podstock unveiled Agent, an AI‑powered assistant embedded in its analytics platform, to automate podcast‑business workflows such as performance reporting, inventory planning, and campaign execution. The tool pulls real‑time data from each organization’s Podstock instance, answering ad‑hoc queries and generating multi‑show...
Spot AI-Generated LinkedIn Posts with 10 Red Flags
Most LinkedIn creators use AI to write their content. How do I know? Because I see the same "AI Fingerprints" ruining what they're trying to say. Here are the 10 biggest AI Red Flags to look out for:
KPMG Has a New Way of Pushing Staff to Make Breakthroughs Using AI — Handing Out Cash Prizes
KPMG has introduced the AI Spark Innovation awards, a cash‑prize program that pays consultants more than the firm’s typical year‑end bonuses for breakthrough AI projects. The initiative, open to staff below the director level, will recognize ideas that generate client...
Research: Using AI Can Stifle Innovation. But It Doesn’t Have To.
A new Management Science study shows that effortless AI‑generated answers boost short‑term productivity but suppress independent exploration, causing firms to converge on a few "good‑enough" solutions. The model predicts, and empirical work confirms, that when reuse is frictionless, the diversity...
Lundbeck Names AI Head in Quest to Become 'Bionic' Company
Lundbeck has promoted Markus Kede to chief AI officer, shifting him from finance to lead the drugmaker's AI agenda. The newly created role reports directly to CEO Charl van Zyl and will be based in Sweden. Kede will oversee AI...
Philosopher David Chalmers: Current AI Interpretability Methods Miss What Matters Most
David Chalmers argues that current AI interpretability focuses on mechanistic analysis and neglects the system's internal beliefs, desires, and intentions. He proposes "propositional interpretability"—a framework that treats AI attitudes like beliefs and goals as observable through "thought logging." Chalmers links...
Most Small Businesses Can’t Afford a Full-Time Finance Chief. So Mastercard Is Debuting a ‘Virtual CFO’ Built with AI
Mastercard is introducing a virtual CFO built on generative AI as the first offering of its “Virtual C‑suite” for small and medium‑sized enterprises. The service will be delivered through banks, accounting platforms and software partners and will provide cash‑flow risk...
CoreWeave’s Flexible AI Cloud Pricing Model Signals Strategic Shift
CoreWeave unveiled a flexible AI‑cloud pricing suite that adds “Flex Reservations” and “Spot” instances to its existing reserved and on‑demand options. The new tiers let customers pay a reduced holding fee for guaranteed capacity ceilings and use‑based rates only when...
AI Networking Startup Eridu Steps Out Of Stealth With More Than $200M In Funding
Eridu, a Saratoga‑based AI networking startup, emerged from stealth with an oversubscribed Series A that raised more than $200 million. The company claims its clean‑sheet switch architecture can cut latency, reduce data‑center power use by up to 70%, and lower capital expenditures...
The Boring Infrastructure That Could Make Agentic AI Happen For Ad Tech
MadConnect has made its Intelligent Connectivity Layer (ICL) generally available, offering a plug‑and‑play infrastructure that links CRMs, CDPs, clean rooms, DSPs and other ad‑tech systems. Built to natively support the open‑source Model Context Protocol (MCP), ICL acts as a switchboard...
The Boring Infrastructure That Could Make Agentic AI Happen For Ad Tech
MadConnect has made its Intelligent Connectivity Layer (ICL) generally available, offering a neutral infrastructure that links CRMs, CDPs, clean rooms, DSPs and other ad‑tech systems for AI agents. The ICL natively supports the open‑source Model Context Protocol (MCP), acting as...
Google Adds Personal Intelligence to Gemini Live
Of course Personal Intelligence makes sense for Gemini Live -> Google is quietly working on a big upgrade for Gemini Live "So, for example, if you have booked a flight but have forgotten the time, you'll be able to use your...
AI Now Writes Most Code, Predictions Proved Spot‑on
Exactly one year ago (10 mar 2025), Dario Amodei: "I think we will be there in 3-6 months, where AI is writing 90% of the code. And then, in 12 months, we may be in a world where AI is writing...
OpenAI Acquires Promptfoo as Enterprise AI Platform Wars Heat Up
OpenAI announced the acquisition of Promptfoo, a fast‑growing platform that automates testing and validation of large‑language‑model prompts. The deal, terms undisclosed, brings Promptfoo’s open‑source tooling and enterprise‑grade safety features under OpenAI’s product umbrella. By integrating Promptfoo, OpenAI aims to streamline...
Terra Portal Adds Human-Governed AI to Live Production Pentesting
Terra Security unveiled Terra Portal, a desktop app that places human‑governed AI at the core of live‑production penetration testing. The tool combines autonomous agents for routine tasks with a Copilot layer for controlled exploitation, shrinking the discovery‑to‑fix timeline from months to...
From POC to Production: Why AI Success Depends on Operational Discipline, Not Just Models
Enterprises can spin up AI proof‑of‑concepts in days, but moving those models into reliable, scalable production remains a major hurdle. The discussion with Deazy and Aveni highlighted that the surrounding system—governance, observability, and cost controls—has become the primary bottleneck, not...
Ericsson, Future Technologies Scale Wireless Infrastructure for Industrial AI
Ericsson and Future Technologies Venture have deepened their 13‑year collaboration to accelerate enterprise wireless and private 5G rollouts for AI‑driven industrial operations across North America. The partnership will combine Ericsson’s cellular technology with Future Technologies’ systems‑integration services, targeting sectors such...
Large Language Models and Medical Misinformation
Large language models are increasingly embedded in clinical documentation, patient chatbots, and medical education, but a new Lancet Digital Health study reveals they can still repeat or partially accept false medical claims. Researchers tested leading models with misinformation‑laden prompts and...
Muscles From Munich? How German Courts Might Stop US Companies From Violating Copyright Through AI Training
The Munich Regional Court heard GEMA’s claim that Suno, a US‑based generative‑music AI, infringed German copyrights by using protected songs for training. The dispute revives the court’s earlier OpenAI ruling but shifts focus to musical arrangements and the cross‑border location...
Enterprise AI Needs Trusted, Proprietary Data Foundations
In a recent interview, Oracle’s Vice President of Development Siva Ravada argues that enterprise AI must rest on trusted, proprietary data foundations. Oracle is weaving geospatial intelligence into its AI‑optimized database, data lakehouse, and core enterprise applications, allowing spatial, graph...
Google AI Voice Models Coming To Google Ads PMax Videos
Google is adding AI‑generated voice‑overs to Performance Max video ads, automatically turning silent creatives into narrated assets. The system pulls text from existing headlines and descriptions, synthesizes natural‑sounding speech, and layers it onto the original video. Advertisers can opt out...
From Human Review to Full Automation: AI’s Oncology Journey
Today, I gave a virtual talk to a leading pharma company's team about the future of oncology. As they asked me to focus on AI's role in that, I'll describe real-life and some hypothetical examples of how the 5 levels...
AI Pulse Exclusive: How Spacely AI Is Bringing Generative AI Into Spatial Design Workflows
Spacely AI delivers a generative AI platform that converts floor plans and rough concepts into realistic 3D visualizations within minutes, targeting architects, interior designers, and real‑estate professionals in the US and Europe. The company embeds AI across its product and...
Broaden AI ROI Metrics to Capture Hidden Benefits
How you define AI's ROI should determine what you measure. IBM suggests that these users widen their search and acknowledge that agentic AI’s ROI should have many components beyond the one or two usually used. It's likely that many of the...
AI Could Disrupt Non‑Partner VC Roles by 2028
Are AI agentic systems going to impact non-Partner VC jobs by 2028 the same way they are in other knowledge-based industries? Poll below. My take in the comments:
Sumo Digital Partners with Arm to Explore AI-Powered "PC-Class" Graphics for Mobile Games
Sumo Digital has partnered with chip designer Arm to evaluate Arm’s new neural technology, which embeds AI‑powered accelerators within its GPUs. The collaboration aims to deliver "PC‑class" graphics on mobile devices by leveraging AI‑driven rendering techniques. Sumo is testing the...
Data Leaders Miss Market Shifts Without Real-Time Tracking
5 questions most data leaders can't answer about their market. Not because they're lazy. Because nobody tracks this. How fast is OpenAI's influence shifting? Who are the real authorities on AI governance? Which entities actually get co-mentioned? Swipe.
Lanús Highlights AI and Agentic Infrastructure at Argentina Week
Glad to be at 🇦🇷 🇺🇸 Argentina Week today in NYC representing Lanús to discuss the future of AI and Agentic Infrastructure https://t.co/XuwDa7qHCm
Proshort Wins Enterprise Adoption with RateGain for Its AI-Native Revenue Platform
Proshort announced that RateGain Travel Technologies has selected its AI‑native revenue intelligence platform for global rollout. The deployment targets higher forecasting accuracy, clearer pipeline visibility, and consistent execution across RateGain’s worldwide sales teams. Proshort’s architecture relies on a proprietary Context...
ChatGPT Leads, but Gemini & Claude Accelerate in AI Race
Both Gemini and Claude have surged -> A look at the top 100 GenAI consumer apps: ChatGPT leads but the race for the “default AI” is on, global usage is splintering by product, and AI agents arrive https://t.co/s0iDDAXa0x https://t.co/gUpqS0Yvrv
Sandboxing AI Coding Agents Prevents Unintended Data Access
AI coding agents running locally can execute commands and access files with the same permissions as the developer — raising the risk of unintended access to sensitive data. Safehouse is a macOS tool that runs agents in a sandbox to limit...
KB Asset Management Launches RISE China AI Semiconductor Top 4 Plus ETF
KB Asset Management and Solactive have launched the RISE China AI Semiconductor Top 4 Plus ETF, which tracks the Solactive China AI Semiconductor Top 4 Plus Index. The index selects 15 Chinese companies listed in Hong Kong or via Stock...
Google AI Learns to Link Out Through New Training Insights
Google's @rmstein on teaching Google AI Search on how to link out, plus more https://t.co/tT09F32Pln interview by @heykahn https://t.co/e5rbkaBzSW
Free AI Tool Finds Verified Experts in Minutes
I built a free AI research desk for journalists, bookers, and podcast producers. It finds verified expert sources in under 3 minutes. No login. No paywall. No pitch inbox. https://t.co/QOpWJ99Sw7
An Industry Benchmark for Data Fairness: Sony’s Alice Xiang
Sony’s AI governance lead Alice Xiang announced FHIBE, a publicly available Fair Human‑centric Image Benchmark that combines ethically sourced, consent‑based data with extensive demographic annotations. The benchmark targets bias measurement in computer‑vision models, addressing the scarcity of responsibly collected datasets....
Google Ads Adds AI Voice to Performance Max Videos
Google AI voice models coming to PMax video campaigns in Google Ads https://t.co/HdEjC1qd7C via @beXtremeArpan https://t.co/K39CiSxc7P
Will Features Even Exist? How AI Is Forcing SaaS To Rethink The Product Itself
A mid‑size SaaS vendor’s top customer is now considering using AI coding tools to build a needed workflow rather than waiting for a traditional feature release. This reflects a broader industry shift where AI‑generated code can deliver narrow, customized functionality...
Sustaining Open Source in the Age of Generative AI
Open source projects are grappling with a surge of AI‑generated pull requests that overwhelm maintainers, prompting Kyverno to publish a concrete AI Usage Policy. The policy acknowledges that while generative AI accelerates code creation, human review capacity remains limited and...
Jensen Huang Warns AI Community's Growing Backlash Anxiety
Notable essay from Jensen Huang. You can feel the AI crowd getting nervous about the backlash https://t.co/HconyGObYw https://t.co/d11coExS4g
Founders Turn “Claudepilled”: Embracing Claude AI Revolution
Why top founders are becoming 'Claudepilled' and what it means Forbes https://t.co/VxLlKcVvKF #AI #ClaudeAI #AIagents https://t.co/Xtot86n70c
Where’s the New Business?
The article introduces the H3 Provocation Framework, a five‑question tool designed to push AI initiatives in commercial real‑estate (CRE) beyond efficiency (H1) and capability (H2) into true transformation (H3). By applying the framework to workflows such as diligence, advisory, occupancy...
The AI Tariff We Are All Going to Pay (And It’s Not Energy Prices)
The article warns that AI’s explosive compute needs are creating an "AI tariff" by monopolizing memory, storage and wafer capacity. DRAM prices have jumped up to 300% and SSD costs have doubled, pushing laptop prices 20‑30% higher. Hyperscalers have locked...
AI Techniques Speed up Forensic Analysis of Crucial Crime Scene Larvae
Researchers at LSU and Texas A&M are using machine‑learning combined with infrared spectroscopy and mass‑spectrometry to identify forensic maggot species, sex, and even toxins within minutes. The approach creates a metabolomic database that can classify insects from chemical fingerprints, eliminating...