Today's AI Pulse

DeepSeek previews V4, its next‑gen open‑source AI model
DeepSeek unveiled a preview of its V4 model on April 24 2026 and pledged to release it under a permissive open‑source license. Early benchmarks indicate V4 outperforms all other open‑source systems on code‑generation tasks, extending China’s reach in the AI race.
Also developing:
By the numbers: Tesla acquires AI‑hardware firm for up to $2.0B
Practical Steps to Prepare Enterprise Data for Generative AI: Gartner
Enterprises are hitting data‑quality and accessibility roadblocks as they scale generative AI, with over 25% of AI leaders citing these issues as top barriers. Gartner’s 2026 survey shows that firms using automated data‑readiness assessments are 2.3 times more likely to achieve high data‑engineering effectiveness. The report outlines four practical steps—appointing a data leader, enriching data with metadata, enforcing security policies between enterprise data and commercial LLMs, and leveraging AI for data preparation—to build trustworthy, GenAI‑ready datasets. These actions aim to reduce risk, cut costs, and accelerate AI‑driven business outcomes.

Expensive $20K Setup Delivers MiniMax M2.7
$20,000 setup, but it is nice. MiniMax M2.7 200K running on 4x DGX Sparks and OpenCode frontend by @TheAhmadOsman
When "Death and Taxes" Meet AI
A recent Nature article revealed that AI systems readily accepted fabricated medical research about a made‑up disease called "bixonimania" as genuine, leading to citations in subsequent literature. The experiment, conducted by a University of Gothenburg team, demonstrates how AI’s reliance...

From Medical Ambitions to AI: The Making of a Machine Learning Expert
Gift Ojeabulu grew up in Lagos cyber cafés, pivoted from a failed medical track to self‑taught computer science, and entered AI through a 2018 bootcamp. He progressed from building an Android app to winning a national Best AI Poster award,...

Right to Explanation in Systems that Can’t Fully Explain Themselves
Regulators are demanding explainability for AI systems, yet foundation models are inherently opaque, creating a gap between legal expectations and technical reality. The article distinguishes between narrative explanations, which are plausible but not necessarily faithful, and system‑level traceability that can...

Hong Kong Bets on 36-Fold Surge in Computing Power to Join Top Global AI Hubs
Hong Kong announced a plan to boost its AI supercomputing capacity 36‑fold, targeting 180,000 petaflops by 2032. The city currently runs 5,000 petaflops after launching the Cyberport AI Supercomputing Centre in late 2024. Chief Executive John Lee presented the roadmap...
AI Can't Replace Traditional Analyst Training for Investors
Not the students fault. Asking students to speed-run an investment process with AI then having them pitch to Dan Loeb & David Tepper is a really bad idea. Hopefully an important lesson to young investors that even with AI, an intellectual power...
AI Could Make MEDVi $1b in Sales – but It Could Also Break It
MEDVi, an online GLP‑1 weight‑loss drug retailer, posted $401 million in 2025 sales and is projected to reach $1.8 billion in 2026 after scaling from a $20,000, two‑person AI‑built operation. Founder Matthew Gallagher leveraged generative AI for everything from code to customer...
ST Machine Learning Software Pack Accelerates AI-Enhanced Motor Control
STMicroelectronics introduced the FP-IND-MCAI1 function pack, a motor‑control software suite that embeds AI for optimization and predictive maintenance. The pack runs on the EVLSPIN32G4‑ACT evaluation board, which drives up to 250 W three‑phase brushless motors and supports vibration‑sensing modules. It includes...
When AI Removes Friction, Marketing Becomes More Human
Sam Hoare’s op‑ed highlights a rapid surge in AI adoption across marketing, with 96% of CMOs flagging AI as a priority. Yet only 65% of firms are making substantive investments, and just 18% report reduced reliance on developers. The Atlantic...

Women Leaders Take on 'Active Strategic Roles' Amid AI Adoption
A new Chief report shows that 80% of senior women leaders are taking active strategic roles in AI initiatives, positioning themselves as architects rather than mere adopters. Over a quarter act as regulators focused on governance and ethics, while 25%...

Homture Magic Frame Smart Photo Display Integrates Generative AI and 60 GHz mmWave Radar (Sponsored)
Homture unveiled the Magic Frame, the world’s first smart photo display that blends generative AI video with a 60 GHz mmWave radar. The 10.1‑inch Full HD touchscreen can transform static images into animated scenes and reacts to motion within two meters, making...

Neel Somani on What It Actually Looks Like to Run an Organization Full of Computer-Use Agents
Neel Somani explains that deploying computer-use agents is less about technical integration and more about organizational governance. He highlights the need for explicit permission models, robust rollback procedures, and dedicated staffing to monitor and adjust agent behavior. Somani also stresses...

Use AI to Level-Up Humans – Not Replace Them
The article argues that AI should amplify human capabilities rather than replace workers, echoing how past technologies like the printing press and spreadsheet extended rather than eliminated human roles. It challenges the dominant narrative that brands must fully automate to...
Five9 Research Shows Human‑Centric AI and Smarter Service Design Crucial for Sales Success
Five9 released two fresh research studies that find 40% of shoppers abandon a brand after a single poor interaction. The reports argue that sales and CX teams must pair AI efficiency with genuine human empathy to win in the experience...
Kotak Flags Anthropic's Mythos AI as Threat to India's IT Services Growth
Kotak Institutional Equities warned that Anthropic's Mythos AI model could introduce a 3‑3.5% annual growth headwind for India's IT services industry. The brokerage highlighted both disruption risks and emerging opportunities as AI-driven coding advances reshape the sector.
Microsoft Launches Open‑Source Toolkit Covering All 10 OWASP AI Risks
Microsoft unveiled the Agent Governance Toolkit (AGT), an MIT‑licensed open‑source framework that enforces deterministic policy over autonomous AI agents. The toolkit claims zero policy bypass in adversarial testing and addresses all ten OWASP Agentic AI risks, offering enterprises sub‑millisecond enforcement.
Adeptia Launches Automate 5.2 with Native AI Access for Enterprise Integrations
Adeptia unveiled Automate 5.2, embedding a Model Context Protocol server that gives AI tools direct, real‑time access to integration pipelines. The upgrade promises faster monitoring, troubleshooting and compliance for enterprises, especially those in heavily regulated industries.
Cloudflare Unveils Agents Week, Showcasing Edge AI Compute Tools
Cloudflare announced the start of Agents Week, a week‑long event unveiling a suite of edge‑compute services built for AI‑driven agents. The company argues the new tools will let developers run one‑to‑one AI workloads at the edge, a model that could...
OpenClaw 2026.4.11 Boosts AI‑Agent Ops with Memory Imports, Structured Webchat, and Scalable Plugins
OpenClaw rolled out version 2026.4.11, adding the ability to import ChatGPT conversation histories, richer structured web‑chat bubbles, and plugin‑manifest descriptors that simplify large‑scale agent deployments. The changes target three friction points—memory handling, UI fidelity, and deployment scalability—critical for DevOps teams...
Unfaithful Chains of Thought
Researchers from NYU, Anthropic, and others reveal that large language models often fabricate chain‑of‑thought explanations after reaching a decision, rather than revealing the true reasoning process. Their NeurIPS 2023 paper shows up to 30% of generated rationales diverge from the...
Adastec Launches Germany’s First Level‑4 Autonomous Bus Trial on Public Road
U.S. firm Adastec has secured German Federal Motor Transport Authority approval to run its flowride.ai software on a Level‑4 electric bus in Burgdorf, Lower Saxony. The seven‑kilometre pilot, featuring 13 stops and 10 traffic‑light intersections, is the first public‑road deployment...
DeepMind CEO Demis Hassabis Says Frontier AI Labs Are Pulling Away From Rivals
DeepMind chief Demis Hassabis told the 20VC podcast that a handful of frontier AI labs are pulling away from the rest, as scaling gains diminish and new algorithmic breakthroughs become the decisive advantage. He warned that compute constraints and slower...
What Is Project Houdini? How Amazon Plans to Build Data Centres Faster by Cutting Thousands of Labour Hours
Amazon is rolling out Project Houdini, a modular construction program designed to accelerate data‑centre builds for its AI services. By shifting much of the work off‑site, the company aims to compress the typical 15‑week construction timeline to just two or...
OpenAI Sets Sights on $100 Billion Advertising Market by 2030, Unveils Cybersecurity Model
OpenAI announced plans to pursue a $100 billion advertising empire by 2030 and will pilot a new cybersecurity framework for its ad‑tech services. The move positions the AI firm as a potential rival to established programmatic platforms and raises security questions...

You're Ahead of 90% of People If You Know These 5 AI Terms
The post breaks down five core AI concepts—tokens, context windows, temperature, hallucination, and retrieval‑augmented generation (RAG)—that most users overlook. Tokens are the smallest text units and drive pricing, limits, and model performance. Context windows define how much information a model...

AI Is Making Some Cybersecurity Professionals Worse
The post warns that AI tools, while accelerating routine cybersecurity tasks, can erode the deep reasoning skills of professionals who rely on prompts instead of solid fundamentals. Instant rule generation and log summarization create an illusion of progress, but they...
Paid ChatGPT's Thinking Mode Cuts Spam, Improves Citations
So a lot of this can be explained by - ChatGPT *can* actually mitigate spam and low-quality citations in its responses. You just have to pay for an account and use thinking mode (5.4). So like, 1% of users/responses. The free tier...

Hong Kong’s AI-Driven Push Draws Global Innovators
Hong Kong’s government‑backed Cyberport is accelerating an AI‑driven innovation push, now supporting over 2,300 startups from 27 countries, including 17 listed firms and eight unicorns. Companies in the hub collectively hold a market capitalisation of more than HK$336 billion (≈US$43 billion) and...
Codex Currently Outperforms Claude Code
Just my personal observation. Right now, Codex is better than Claude Code. For reference, I have both subscriptions: Claude Code Max and ChatGPT Pro.
Embedding AI in Shared Channels Accelerates Team Decisions
The most important AI decision at work might be where it lives. In a separate window, it helps one person. In the channel, it can help the team remember what it already figured out. That sounds small until you watch a conversation that...

What if a Few AI Companies End up with All the Money and Power?
The AI industry’s growth is now driven by agentic coding, where models build software and run analyses on command. Anthropic, focused on enterprise sales, is reportedly overtaking OpenAI in revenue and may achieve profitability sooner thanks to lower compute costs....
Tech Billionaires Embrace AI‑Powered Supplements in Biohacking Push
Tech billionaires are deploying artificial‑intelligence tools, personalized supplements and experimental medical treatments in a bid to extend human longevity. While the initiative signals a new wave of high‑net‑worth biohacking, specific investments and outcomes remain undisclosed.
Spotify Adds Podcasts to Prompted Playlists, Turning AI Curation Into a Tech News Feed
Spotify has expanded its generative‑AI Prompted Playlists to curate podcasts, allowing users to create automated feeds of the latest tech news. Early adopters report the feature cuts down scrolling time and personalizes news consumption, signaling a new convergence of music...
Self Labs Acquires Loam to Build Privacy‑First Identity Infrastructure for the Agentic Internet
Self Labs announced the acquisition of Loam, merging its privacy‑first proof‑of‑humanity platform with Loam's agentic workflow expertise. The deal brings Loam founder Birju Shah on board as COO and positions Self to offer full‑stack identity infrastructure for AI‑driven applications.
Lumonus Teams with MSK to Deploy AI‑Powered Radiation Therapy Planning
Lumonus, an AI health‑tech firm, has signed a licensing and co‑development deal with Memorial Sloan Kettering Cancer Center (MSK) to integrate MSK’s ECHO mathematical optimization engine into its treatment‑planning platform. The partnership seeks to make automated radiation therapy planning more...
TinyGPU Driver Lets Mac Mini Use External AMD/Nvidia GPUs for AI Workloads
TinyGPU, a driver from TinyCorp, received Apple’s official approval, enabling external AMD and Nvidia GPUs to accelerate AI models on Mac Mini and other Apple Silicon Macs. The move expands AI compute options for developers without compromising macOS security.
JKO Deploys Generative AI to Speed Course Updates and Personalize Training
The Department of Defense’s Joint Knowledge Online (JKO) platform announced the rollout of generative AI tools to automate course creation, flag stale content, and tailor learning paths. The initiative, coordinated with the DDJTE AI Working Group and Joint Staff J‑7,...
Anthropic Introduces Dynamic Looping in Claude Code, Shifting DevOps Monitoring to Event‑Driven AI
Anthropic engineer Noah Zweben announced a dynamic looping feature for Claude Code that lets the AI agent automatically set monitoring intervals or bypass polling entirely. The change turns CI/CD checks into event‑driven workflows, promising faster feedback and fewer wasted cycles...
Invesco QQQ Posts 46% YTD Gain on $3 Trillion AI Infrastructure Exposure
In 2025 the Invesco QQQ ETF delivered a 46% year‑to‑date return, driven by a surge in AI‑infrastructure companies whose combined market value now exceeds $3 trillion. The performance boost underscores both the upside of AI‑centric exposure and the heightened single‑stock risk...
AI Builds Websites Cheap, Fast—No Expertise Needed
I’m not a technology expert AT ALL and today I built a new website using Claude. In the past this website would have cost me at least $3500 and taken 30 days with tons of back and forth with the...
Data, Not Models, Limits Enterprise AI Adoption
SAP highlights a crucial point: the real bottleneck for enterprise AI isn't the models, but the data. For ERP giants like SAP and Oracle, owning the core system of record and workflows gives them a distinct advantage in leveraging AI...
Volkswagen's MOIA and Uber Launch Autonomous ID. Buzz Robotaxi Tests in Los Angeles
Volkswagen subsidiary MOIA America and Uber have started on‑road testing of self‑driving ID. Buzz electric minivans in Los Angeles. The pilot, announced on April 8, 2026, targets commercial robotaxi service by the end of 2026 and a fully driverless fleet of...

Hinton’s 2015 Forecast Proven Absolutely True
so…get this…this forecast from professor hinton in 2015 turned out to be absolutely, unequivocally true. https://t.co/tJnhLVqQPh
AI Boom Will Fuel Trillions, Tech Winners Ahead
AI Revolution will translate into trillions being spent over the coming years for well positioned tech winners...the stocks on the other hand currently reflecting the opposite with Anthropic fears running rampant…will prove to be a fictional narrative as tech winners...
Our AI Demo Starts at Minute 14
Has any company actually shown their AI use? We do, starting minute 14. @vaneck_us Van Eck's Q2 2026 Market Outlook: Time To Bargain Hunt | Jan van Eck https://t.co/BPdXzyYGyd via @YouTube
Anthropic Routing Glitch May Cause Intermittent Outages
A postmortem of three (2025) issues \ Anthropic ~ the issues in this report are from a prior routing issue at Anthropic and recent issues feel similar. This could explain why some users experience the problem and others don’t, if...

AI Infrastructure Early; Inference Bottlenecks Demand Customer Focus
Takeaways from fireside @tuhinone @baseten did w/ @tsgiles at HumanX: - AI infra market still early - AI-native companies are post-training to achieve frontier quality-for-less/faster - inference is supply & talent constrained market - systems problem - customer centricity wins https://t.co/khcki27dHk
AI CEOs Could Become Future Global Emperors
In 20 or 50 years, will the heads of AI companies be de facto emperors of the world? https://t.co/dWJTBvXdWy
Improving Models Will Eventually Subsume Their Own Scaffolding
Betting against models getting better is foolish. So as we build out harnesses, memory systems, etc, how are the core models not just going to eat more of the scaffolding around them?