Import AI 446: Nuclear LLMs; China's Big AI Benchmark; Measurement and AI Policy
This episode explores how measurement drives AI governance, highlighting Jacob Steinhardt's argument that better metrics can lower policy compliance costs and shape incentives, much like CO₂ monitoring or COVID testing. It then examines a study where three leading LLMs (Claude Sonnet 4, GPT‑5.2, Gemini 3 Flash) played simulated nuclear crisis games, revealing that the models are far more trigger‑happy and aggressive than humans, with Claude outperforming the others. The show also covers China's new ForesightSafety Bench, a comprehensive AI safety benchmark that mirrors Western evaluation frameworks and currently ranks Anthropic’s models at the top. Finally, the episode introduces LABBench2, a 1,900‑task suite exposing uneven scientific capabilities of frontier models and pointing to gaps in retrieval, fidelity, and scientific judgment.
New Browser API WebMCP : Actually Makes AI Agents Work Efficiently
WebMCP, a joint Google‑Microsoft browser API, enables websites to expose structured actions that AI agents can invoke directly, eliminating the need for HTML parsing or screen scraping. By acting as mini‑MCP servers, sites provide declarative tools—often defined via JavaScript or...
Astera Labs Acquired Pliops
Astera Labs announced the acquisition of Pliops, paving the way for its first research and development center in Israel. The move, confirmed by former Pliops CRO Marius Tudor, expands Astera’s global footprint and aligns with its strategy to accelerate AI‑connectivity...
Crusoe Launches Command Center: A Unified Operations Platform for High-Performance AI Workloads
Crusoe unveiled Command Center, a unified operations platform that consolidates GPU telemetry, logging, and orchestration for high‑performance AI workloads. The solution offers out‑of‑the‑box metrics, custom metric ingestion, and topology‑aware views, while integrating tightly with Crusoe Managed Kubernetes, Slurm, and AutoClusters....
Ax3.ai Appoints Marius Tudor as Chief Revenue Officer to Accelerate Global Growth
Ax3.ai appointed Marius Tudor as Chief Revenue Officer to lead its global go‑to‑market strategy. Tudor brings more than 25 years of technical sales experience in infrastructure and hardware. He will expand direct sales channels, strengthen the partner ecosystem, and target...
Meet ASKB: A First Look at the Future of the Bloomberg Terminal in the Age of Agentic AI
Bloomberg unveiled ASKB, a conversational AI interface now in beta on the Bloomberg Terminal, aiming to accelerate investment research. The agentic system taps Bloomberg’s massive data, news, and research libraries, delivering answers with transparent source attribution and even the underlying...
LLMs Still Struggle With Medical Misinformation
A Lancet Digital Health study evaluated 20 large language models with over three million medical prompts, revealing a high susceptibility to misinformation. Neutral prompts led to a 32% acceptance rate of false information, which rose to 46% when embedded in...
NotebookLM for Lawyers: A Small Hammer for Big Document Problems
Google’s NotebookLM, an AI‑powered private notebook, lets lawyers upload up to 300 documents per notebook and query them with citation‑backed answers. The tool excels at digesting litigation files, clustering discovery material, and generating timelines, briefs, and cross‑examination outlines without pulling...
How Much Is AI Hurting Entry-Level Hiring?
A Stanford working paper reports a 16% employment decline for 22‑ to 25‑year‑olds in AI‑exposed roles such as software engineering and customer service since ChatGPT’s release. The findings have become a reference point in the AI‑jobs debate. However, two senior...
Survey: Only 17% Trust AI Without Oversight
The Connext Global 2026 AI Oversight Survey of 1,000 U.S. workers shows that just 17 % trust AI to operate without human supervision, while 70 % rely on a human safety net—either light review or dedicated oversight. Monitoring is pervasive, with 82 %...
The Post-Truth Purgatory
By 2026, up to 90 % of online content is expected to be synthetically generated, thrusting the legal profession into a crisis of evidentiary authenticity. Courts are grappling with deepfakes and AI‑hallucinated citations, forcing lawyers to prove not just that an...
Telefónica, Mavenir Partner to Drive AI Innovation in Telecom
Telefónica and Mavenir have signed an MOU to launch an AI Innovation Hub that will serve as a live testbed for autonomous core network technologies. The hub will enable development and validation of AI‑driven orchestration, intent‑based services, and monetization frameworks...
AMD VEK385 Versal AI Edge Gen 2 FPGA Evaluation Kit Plugs Directly Into a PCIe Gen5/Gen4 Slot
AMD unveiled the VEK385 evaluation kit built around the Versal AI Edge Gen 2 XC2VE3858 SoC FPGA. The board combines eight Cortex‑A78AE cores, ten Cortex‑R52 cores, a Mali‑G78AE GPU, 144 AI Engine‑ML v2 tiles delivering up to 184 INT8 TOPS, and 20 GB...
The BYOAI Trap
The 2026 L&D Global Sentiment Survey reveals that AI interest has peaked, yet practitioners are increasingly using personal, unmanaged AI tools to meet “doing more with less” pressures. Over 40,000 words of respondent commentary highlight rising concerns about redundancies and...
NVIDIA’s Blackwell Ultra GB300 Sets New Standards in Long-Context Inference and Significantly Outperforms GB200
NVIDIA unveiled the Blackwell Ultra GB300 NVL72 rack, targeting long‑context inference for agentic AI models. Benchmarks show a 1.4‑1.5× performance uplift over the GB200, with a peak throughput of 226.2 tokens per second and latency gains of roughly 1.58×. The gains...
I Switched Everything to Local AI and Stopped Sending My Documents to the Cloud
A tech writer realized that uploading confidential documents to cloud‑based AI services violated data‑privacy expectations and switched to a fully local solution. After reviewing terms of service, they adopted AnythingLLM, an open‑source desktop application that runs AI models entirely on...
Intapp to Partner with Harvey Bringing Ethical Wall Enforcement Directly Into the Platform
Intapp and Harvey announced a strategic partnership to embed Intapp Walls for AI directly into Harvey’s generative AI platform for legal professionals. The integration will automatically synchronize Intapp’s ethical‑wall policies with Harvey’s Assistant, Vault, and Workflows modules, ensuring AI‑driven work...
How to Calculate ROI for AI Investments (2026) – Shopify
The episode breaks down how ecommerce businesses can accurately calculate AI ROI by defining a clear formula, accounting for both direct financial gains and indirect benefits such as risk mitigation, operational efficiencies, and data‑related costs. It highlights why AI ROI...
SKT 6G ATHENA White Paper: A Mid-to-Long Term Network Evolution Strategy for the AI Era
SK Telecom released the ATHENA white paper outlining its mid‑to‑long‑term strategy for a 6G network that is AI‑native, Zero‑Trust, and open. The framework, encapsulated by the ATHENA acronym, maps out how AI will be embedded across radio, core, transport and...
AI/ML, Multiscale Modeling, and Emergence
Artificial intelligence and machine learning are rapidly reshaping materials design, with major tech firms and startups pursuing inverse‑design platforms that translate target properties into synthesizable compounds. Recent reviews highlight efforts from Google DeepMind, Microsoft, Meta, Toyota Research Institute, IBM and...
Beyond Capacity: Why AI Is Forcing a Building-Level Performance Reckoning
AI‑driven workloads have accelerated U.S. electricity demand, with data centers responsible for roughly 60 % of last year’s load increase and a 150 % rise over five years. This non‑linear growth exposes a hidden vulnerability at the building level, where inefficiencies in...
The Long-Term Reality of Hyperscalers: The Good, The Bad, and The Ugly Scenarios
OpenAI disclosed that its adjusted gross margin on inference dropped to 33% in 2025, down from 40% the previous year, as inference costs quadrupled. The company also projected training expenditures to rise sharply, reaching $32 billion this year and $65 billion next...
From the Microsoft Dynamics 365 Finance and Supply Chain Management: Preparing Data for AI; Secure Email Setup; Connect Power BI;...
Microsoft’s Dynamics 365 Finance and Supply Chain Management ecosystem is evolving to meet AI‑driven demands, with recent blog posts outlining how firms can ready ERP data for artificial intelligence, configure secure SMTP email directly within the platform, and link Power BI for...
INL Partners with NVIDIA on Nuclear AI Apps
The Idaho National Laboratory (INL) has teamed up with NVIDIA to embed artificial intelligence across the nuclear reactor lifecycle, from design to operation. Leveraging NVIDIA’s GPU‑accelerated computing, the partnership aims to double deployment schedules and cut operational costs by more...
AI Prompting for Legal Professionals
The article outlines how legal professionals can harness generative AI by treating prompts like legal questions, emphasizing that vague inputs produce useless outputs. It introduces the 7 Ps Framework—persona, product, prompt, purpose, prime, privacy, and polish—as a systematic method for crafting...
Waymo’s Shocking Data & Uber’s Infrastructure Pivot
Waymo is operating roughly 3,000 autonomous vehicles with only 70 remote‑assistance agents, a ratio that analysts tout as evidence of a technology lead. About half of those agents are outsourced to the Philippines, raising political and operational concerns. Uber announced...
Impressive Ride With Tesla FSD in Seville’s Narrow Old Town.
A LinkedIn video shows a Tesla equipped with Full Self‑Driving navigating Seville’s narrow old‑town streets without any safety‑driver assistance. The car performs two reverse maneuvers to negotiate a tight corner while pedestrians and bollards line the path. The footage highlights...
Things Are Getting Wild: Re-Tool Everything for Speed
The author warns that AI is reshaping cybersecurity, creating a tidal wave of new software‑generated vulnerabilities while simultaneously giving attackers tools to industrialize exploits. Simultaneously, AI‑generated content erodes trust, making authenticity a critical challenge. Enterprises must build a robust agentic...
OK26: 5 AI Cold Call Training Scenarios Every Outbound SDR Team Should Run
Troy Johnson, SDR Enablement Program Manager at JumpCloud, explains how his team leverages AI‑driven role‑play to certify outbound SDRs before they make live calls. The program includes more than 100 practice scenarios, a March‑Madness‑style competition, and monthly updates to talk...
WordPress, AI, Plugins, Future of Software Engineering
The post outlines how AI is reshaping the WordPress ecosystem, from a flood of AI‑generated plugins that introduce new security risks to the need for large‑scale audit infrastructure. It advises agencies to pivot from billable hours to outcome‑based pricing, leveraging...
Science Corp. And Neurosoft Bioelectronics Announce Novel BCI Ecosystem Partnership
Science Corporation and Neurosoft Bioelectronics have signed a multiyear, multimillion‑dollar partnership granting Neurosoft access to Science’s full‑stack clinical‑grade BCI tools, known as the Science BCI Ecosystem. The deal lets Neurosoft develop minimally invasive, fully implantable brain‑computer interfaces for tinnitus, epilepsy...
WestFax Launches AI-Powered Intelligent Document Processing Platform for Healthcare
WestFax launched Comprehend, an AI‑powered Intelligent Document Processing platform for healthcare, now available across all its service tiers. The solution uses OCR, AI and FHIR‑aligned models to convert inbound fax, email and file‑based documents into searchable PDFs, classify types, and...
Ericsson Goes with Custom Silicon (Rather than Nvidia GPUs) for AI RAN
Ahead of MWC Barcelona 2026 Ericsson unveiled its AI‑RAN portfolio, opting for purpose‑built ASIC silicon instead of Nvidia GPUs and confirming Intel as its only commercial silicon partner. The company highlighted a software‑defined approach that uses hardware abstraction layers to...
AI Silos: The New Data Fragmentation Problem Inside Law Firms
Law firms are rapidly adopting a patchwork of AI applications, but the lack of a unified architecture is creating isolated data silos. These "AI silos" prevent seamless knowledge sharing across practice groups and hinder the firm’s ability to leverage collective...
Emerson Redefines Scalable Reliability with AI-Driven APM
Emerson’s Aspen Technology unveiled the latest version of its Mtell asset performance management platform on Jan. 22, adding AI-driven capabilities that extend from basic health monitoring to predictive failure analytics. The upgrade features industry‑specific templates, AI‑powered alert grouping, direct integration with...
Napster, Scaling Up Launch AI-Powered Coaching
Napster, an embodied AI firm, announced a partnership with Scaling Up to embed its AI platform into the global executive‑coaching network. The collaboration will generate AI‑driven digital twins of Scaling Up’s seasoned coaches, delivering the proven growth framework 24/7 in multiple languages....
AI Tool Sets New Standard in Diagnosing Rare Diseases
A new multi‑agent system called DeepRare, built on the DeepSeek‑V3 large language model and over 40 specialized tools, outperformed 15 competing AI models and human physicians in diagnosing rare diseases. Across 6,401 cases covering 2,919 rare conditions, it achieved a...
AI in Edu: Signals & Moves (Feb 20, 2026)
In this episode Tom outlines three emerging signals in AI for education: the rise of agentic AI that can complete LMS tasks, the shift from optional chatbots to embedded AI feedback within grading workflows, and growing concerns about platform fatigue...
IBM Introduces Autonomous Storage with New FlashSystem Portfolio Powered by Agentic AI
IBM unveiled three new FlashSystem models—5600, 7600 and 9600—featuring the FlashSystem.ai suite and a 5th‑Gen FlashCore module. The AI‑driven services claim up to 90% reduction in manual storage management and up to 57% cost savings on the flagship 9600. Hardware‑accelerated...
Cisco Unveiled Silicon One G300, Advanced Systems and Optics to Power and Scale AI Data Centers for Agentic Era
Cisco unveiled its Silicon One G300 switch silicon, delivering 102.4 Tb/s of bandwidth to power gigawatt‑scale AI clusters. The G300 powers new N9000 and 8000 systems that use 100% liquid cooling and 1.6 Tb/s optics, boosting energy efficiency by up to 70% and...
7 XGBoost Tricks for More Accurate Predictive Models
The article outlines seven practical XGBoost tricks that boost predictive accuracy on tabular data. It demonstrates how adjusting learning rate, tree depth, subsampling, regularization, early stopping, hyper‑parameter search, and class weighting can transform a baseline model. Code snippets using the...
DesignCon 2026: Keysight to Showcase Advanced AI Data Center and High-Speed Interconnect Validation
Keysight Technologies will present end‑to‑end AI data‑center and high‑speed interconnect validation solutions at DesignCon 2026 in Santa Clara. The showcase includes a 3D chiplet design workflow, signal‑integrity testing up to 3.2 Tbps, DDR5 and GDDR7 memory validation, and PCIe 7.0 PAM4 performance verification....
Bridging AI and Sports Science: How Model Context Protocols (MCPs) and Retrieval-Augmented Generation (RAG ) Systems Can Personalize Training
Model Context Protocols (MCPs) provide a standardized bridge that lets large language models pull live athlete data from platforms like AthleteSR, Strava, or Garmin. Retrieval‑Augmented Generation (RAG) layers sport‑science knowledge from textbooks and research into the model’s output, reducing hallucinations....
How AI Found Better Battery Materials Among 14 Million Possibilities
A collaboration between McGill University, Mila‑Quebec, and Université de Montréal built a closed‑loop system that couples high‑throughput robotic synthesis with multi‑objective Bayesian optimization to explore roughly 14.2 million triple‑doped LiCoPO₄ cathode compositions. Using a set‑transformer surrogate and a multi‑task Gaussian process,...
Nobody Gets Sued but the Doctor: The Legal Vacuum at the Center of the AI Physician Revolution
The episode examines the legal vacuum surrounding AI‑assisted clinical decision‑making, highlighting that while the FDA has cleared over 1,300 AI medical devices, adoption remains low and physicians bear virtually all malpractice liability. Data shows a rapid rise in AI use...
Entanglement Boosts Machine Learning of Quantum Systems
Researchers at Universitat Autònoma de Barcelona introduced an entanglement‑assisted learning strategy for quantum generative adversarial networks (QGANs). By coupling a randomly initialized auxiliary qubit during training, the method overcomes plateaus and local minima that limit QGAN scalability. In a 72‑qubit...
16 Air Assault Demos AI-Enabled Air Manoeuvre Strike
The British Army’s 16 Air Assault Brigade Combat Team showcased AI‑enabled reconnaissance‑strike capabilities during a visit by Chief of the General Staff General Sir Roly Walker. The demonstration linked new hardware, software and procedures to accelerate target acquisition and air‑delivered...
Organizational Readiness, Not Employee Resistance to Change, Is the Biggest Barrier to AI Success
New research from Qualtrics and Celonis shows that employee resistance is no longer the primary obstacle to AI adoption; instead, organizational readiness is. While 52% of workers regularly use AI and 89% of leaders view AI as a competitive opportunity,...
We Tried Using AI for Real Building Work. Here Is What Actually Happened.
At AHR Expo 2026, SES Consulting’s Brad White and Cochrane Supply’s Jacob Fenley tested generative AI on real building data. ChatGPT matched 80% of recommissioning measures but hallucinated equipment specs, while Gemini accurately extracted data from photos, PDFs, and handwritten...
It’s Time for an End to Two-Tier Tech in the NHS
The NHS is rapidly adopting AI and patient‑facing tools, yet back‑office staff such as roster managers remain stuck with outdated systems. This digital divide fuels stress, mismatched rotas, and higher temporary‑staff costs, contributing to the wider retention crisis. Trusts that...
