Pulse on AI

Daily AI roundup: Google launches Gemini 3 and Antigravity coding, DeepMind’s WeatherNext 2 speeds multi-scenario forecasts, Microsoft + NVIDIA plan a $15B stake in Anthropic, and Microsoft Research Africa debuts Project Gecko for hyper-local, low-cost AI. We discuss Pichai’s bubble and energy warnings, Klarna’s AI-driven productivity and comp shifts, platform engineering for gen AI, fragmented global AI regulation, Roblox’s child-safety challenges, Linus Torvalds on vibe coding, xAI Grok 4.1’s leaderboard wins, new research in distributional RL, and a surge in African cybersecurity breaches. Three takeaways: build dependable AI with platform engineering, align on ROI and energy efficiency, and treat safety and inclusion as core features.

Sources:

• Elon Musk’s Grok 4.1 Is the Best AI Model on LMArena Text
• Google CEO: If an AI bubble pops, no one is getting out clean
• Accelerated Distributional Temporal Difference Learning with Linear Function Approximation
• Accelerating generative AI applications with a platform engineering approach
• Google launches WeatherNext 2 with FGN architecture
• Roblox: el videojuego donde millones de menores se cruzan con adultos desconocidos y que preocupa a padres, pediatras y fiscales
• Linus Torvalds is OK with vibe coding as long as it's not used for anything that matters
• How To Raise Capital When You Don’t Sound Like An Insider
• 👨🏿‍🚀TechCabal Daily – Jumia eyes new markets
• Google launches Gemini 3 with new coding app and record benchmark scores
• Is the AI bubble about to burst? What to watch for as the markets wobble
• KI-Regulierung weltweit: Die Strategien anderer Länder
• Klarna says AI drive has helped halve staff numbers and boost pay
• Best Black Friday smartwatch and fitness tracker deals 2025: Top sales out early

Cloudflare’s global hiccup took X down and showed why redundancy matters; ORCA found LLMs still flub real-world math without tool use; Ex Machina reignited debates on AI creativity and autonomy; parasocial relationships went mainstream and 2wai’s grief avatars sparked ethical alarms; Bexorg is scaling an AI-plus-human-brain platform for CNS drug discovery; EXL is betting on the data and fine-tuning layer over GPUs; OnePlus 15’s great AI features meet awkward defaults and phantom touches; robots learned faster with imitation and simulation while Amazon urged pragmatism; new research generalized the BBP transition for PCA under sparse noise; Dealism raised to build AI sales agents; Microsoft pushed a ‘positive-sum’ AI vision and agent pricing; the EU AI Act’s first phase hit GPAI providers with lifecycle obligations; and a dev shipped an iOS app in 3 days using AI-assisted ‘vibe coding.’

Sources:

• Bexorg Appoints Sean Murphy, Ph.D., as Chief Technology Officer
• Kapoor’s Curiosity for Context Keeps EXL Focused on Data
• PCA recovery thresholds in low-rank matrix inference with sparse noise
• Is X (Twitter) down? Why your timeline isn’t loading
• ¿Robots cuadrúpedos, humanoides o industriales? De Fukushima a los almacenes de Amazon, las máquinas trabajadoras conquistan el mundo físico
• AI is actually bad at math, ORCA shows
• Ex-Youzan COO's AI Startup Dealism Raises $15M from Hillhouse, Sequoia China
• Ex Machina: could “superintelligence” challenge the idea of creativity as a uniquely human activity?
• I Built an IOS App in 3 Days with Literally No Prior Swift Knowledge
• KI-Modelle nach den Vorgaben des AI Acts entwerfen
• Feel a connection to a celebrity you don’t know? There’s a word for that
• Disney star debuts AI avatars of the dead
• ChatGPT no funciona: cómo comprobar si está caído o es un problema tuyo
• I uncovered the OnePlus 15's flaws, but this unexpected design feature is the real problem

A 30-minute deep dive into the latest across AI: a new structure-aware SAT encoding breakthrough for abstract argumentation that preserves clique-width; Xiaomi’s safety-first EV philosophy with autonomy implications; SeaPal’s AI fish tank for empathy-driven early education; the surge of AI-chatbot “infidelity” cases and the ethics around them; Yann LeCun vs. Anthropic on AI regulation and open-source competition; how cities use AI to fix potholes and guardrails while safeguarding privacy; a case for U.S. open-source AI leadership; market jitters around AI valuations; KubeCon’s cloud-native security updates and managing AI agent identities; Oracle’s Multicloud Universal Credits and what they mean for AI workload portability; a sober look at Tesla’s robotaxi and humanoid AI milestones; and the most common—and avoidable—mistakes companies make integrating AI/ML, from data foundations to A/B testing and MLOps.

Sources:

• Oracle Just Solved Multicloud’s Biggest Problem
• Structure-Aware Encodings of Argumentation Properties for Clique-width
• Why Elon Musk won't ever realize the shareholder-approved Tesla payout
• Lei Jun Says Xiaomi EVs Are Built to the Industry’s Highest Safety Standards
• Is the OnePlus 15 the best $900 Android phone? Here’s how it compares
• Databricks co-founder argues US must go open source to beat China in AI
• Fund at First Pitch finale at XIN Summit: a warm, child-focused AI interactive fish tank for early education
• Cities and states are turning to AI to improve road safety
• LeCun accuses Anthropic of exploiting AI cyberattack fears for regulatory capture
• Critical Mistakes Companies Make When Integrating AI/ML into Their Processes
• KubeCon: Fortschritte, aber auch mehr Arbeit bei Cloud-Native und Cybersecurity
• Global markets struggle after tech sell-off and fears over Chinese economy
• Hay personas poniendo los cuernos a sus parejas. Todo normal excepto porque lo están haciendo con una IA

OpenAI rolls out GPT-5.1 with adaptive reasoning, extended prompt caching, and new coding tools, while also fixing ChatGPT’s overuse of em dashes via Custom Instructions. A researcher quantifies how similar LLM outputs are and highlights worldview gaps. A workshop aims to bridge logic-based reasoning with transformers. In the enterprise, a cloud-sales leader illustrates how AI demand meets cloud scale; WisdomAI raises $50M to push agentic analytics; Microsoft taps OpenAI’s custom chip designs to accelerate its silicon strategy; and Deepwatch layoffs reflect workforce shifts toward AI. Pop culture and ethics collide as ElevenLabs licenses celebrity voices—including deceased figures—and an AI act tops Billboard’s country sales chart. Robotics headlines range from Russia’s tumbling humanoid debut to Star CM and Unitree’s IP-themed consumer robots. A Spanish interview with Justo Hidalgo weighs emergent abilities, governance, and the limits of current LLMs. The episode closes with practical takeaways: adaptive AI is here, governance/provenance are essential, and infrastructure determines who scales safely.

Sources:

• WisdomAI Raises $50 Million to Lead the Next Gen AI Analytics and BI
• Are all LLMs the same? I set out to measure their similarity.
• ChatGPT Will Stop Using Em Dashes If Instructed So
• Good Luck, Have Fun, Don’t Die trailer ushers in AI apocalypse
• Proceedings of the Second International Workshop on Next-Generation Language Models for Knowledge Representation and Reasoning (NeLaMKRR 2025)
• Deepwatch layoffs deepen sector trend as firms reallocate to AI
• Justo Hidalgo, doctor en Ciencia de Datos: “Existen capacidades que surgen de la IA a partir de cierta complejidad y son impredecibles”
• Russia’s first autonomous humanoid robot staggers and falls on debut
• Introducing GPT-5.1 for developers
• Star CM and Unitree Form Joint Venture to Develop Consumer IP Robots; Shares Surge Over 20%
• Microsoft’s plan to fix its chip problem is, partly, to let OpenAI do the heavy lifting
• Henry Hill’s journey to the frontlines of cloud and artificial intelligence
• Matthew McConaughey und Michael Caine lizenzieren ihre Stimme für KI-Unternehmen
• Esta canción está número uno en la lista Billboard en EEUU. Y ha sido generada por IA

Max and AI expert Keira Sobol break down: Africa’s first multi-model LLM exchange for telcos; Kenya’s M-Tiba health data breach; Vodacom holding on to M-Pesa; Nigeria’s telecom boom; Vodacom–Starlink LEO partnership; the Probably Approximately Correct (PAC) learning framework and why some problems stay hard; OpenAI’s GPT-5.1 Instant and Thinking with eight preset personalities and adaptive reasoning; leaked reports on OpenAI’s inference spending and revenue signals; Germany’s ruling against ChatGPT on copyright; RECAP, a new method to expose LLM memorization; SoftBank’s $40B bet on OpenAI and selling NVIDIA; AI PCs like Asus ProArt P16; the agent era’s metrics beyond CAC/LTV; Baidu’s Xiaodu AI glasses; Nirmata’s AI Kubernetes policy assistant; the rising value of skilled trades for data center buildouts; and an AI security workshop.

Sources:

• Nirmata Launches AI Platform Engineer
• SoftBank Goes All In on OpenAI, Cashes Out Entire NVIDIA Stake
• OpenAI walks a tricky tightrope with GPT-5.1’s eight new personalities
• The Probably Approximately Correct Learning Model in Computational Learning Theory
• OpenAI’s viability called into question by reported inference spending with Microsoft
• Baidu's Xiaodu AI Glasses Pro Now Available, Priced at 2,299 Yuan
• 👨🏿‍🚀TechCabal Daily – Vodacom says no to M-PESA spin-off
• Court rules that OpenAI violated German copyright law; orders it to pay damages
• New RECAP tool exposes just how much copyrighted text LLM's can regurgitate
• AI Hype: Don’t Overestimate the Impact of AI
• Twilight of the user
• iX-Workshop: KI-Methoden und -Werkzeuge für die IT-Sicherheit
• Jensen Huang, CEO de NVIDIA: "Los verdaderos ganadores de la carrera de la IA serán los electricistas o fontaneros"
• My favorite ultraportable laptop has MacBook Pro written all over it (but isn't made by Apple)

Today’s episode dives into practical shifts and structural realities shaping AI. We cover the push to unify sales workflows and prioritize rep effectiveness over tool sprawl; the UK’s plan to pre-test AI models for child-safety risks; ESET’s RMM integration for MSPs; OpenAI’s capital crunch versus broader CHIPS Act tax credits; Apple reserving over half of TSMC’s 2nm capacity and the ripple effects on AI compute; reports that Meta’s Yann LeCun plans a world-models startup; Google’s €5.5B AI data center investment in Germany and sustainability scrutiny; advanced feature engineering methods for high-stakes models; rising calls for secret key hygiene; Google’s LearnLM RCT in math tutoring and new education funding; research showing knowledge edits often decay after fine-tuning and that memorization can be separated from reasoning pathways (with math tied to memory); licensed AI voice marketplaces and improved transcription; evolving copyright and attribution norms; Samsung’s ambient AI; Google’s privacy-hardened cloud AI; and KPIT’s momentum in AI-defined vehicles. Three takeaways: prioritize effectiveness, embed safety and governance, and remember AI progress hinges on real-world infrastructure.

Sources:

• ESET to aid MSPs by integrating with ConnectWise Asio
• KPIT Reports 21st Consecutive Quarterly Growth with $232 Mn in New Deals
• Researchers isolate memorization from reasoning in AI neural networks
• Quantifying Edits Decay in Fine-tuned LLMs
• Our latest commitments in AI and learning
• Apple reserves over 50% of TSMC 2nm capacity for 2026
• Altman sticks a different hand out, wants tax credits instead of gov loans
• Expert-Level Feature Engineering: Advanced Techniques for High-Stakes Models
• UK plans pre-release AI testing to prevent child abuse imagery
• Leaks auf Github: Top-KI-Unternehmen haben ihre Keys nicht im Griff
• 5,5 Milliarden Euro: Googles "BIP-Booster" für KI in Deutschland
• Meta's chief AI scientist maps his exit
• OpenAI va a tener que pagar una fortuna en obligaciones crediticias en 2026. Hoy por hoy las cuentas no salen
• The secret to improving your sales team's success rate isn't more tools and data

Today’s episode explains Nigeria’s landmark bill elevating NITDA as a digital super‑regulator, the strategic implications of Intel’s AI chief joining OpenAI, Microsoft’s “Whisper Leak” side‑channel risk to AI chat privacy, and why the AI boom resembles the dotcom era with crucial differences. We cover the rise of private AI accelerators in India, surging AI healthcare investment, pragmatic LLM evaluation methods, a theory paper on echo‑state networks’ memory bias, Sam Altman’s take on AI poetry versus human provenance, and transparency concerns in Europol’s partnerships with US surveillance tech. Three takeaways: clarify governance, focus on unit economics, and build for trust.

Sources:

• Ondaro Names New CFO and Elevates Key Leaders
• As Public Support for AI Startups Lags, Private Accelerators Take the Lead
• NASA is kind of a mess: Here are the top priorities for a new administrator
• Generalization in Representation Models via Random Matrix Theory: Application to Recurrent Networks
• Intel CTO and AI boss quits to join OpenAI after just six months in the job
• Everything You Need to Know About LLM Evaluation Metrics
• Crunchbase Sector Snapshot: Funding To AI-Related Healthcare Startups Is Robust This Year
• NITDA emerges as Nigeria’s digital super-regulator under new bill
• ‘Breaking Bad’ creator’s new show ‘Pluribus’ was emphatically ‘made by humans,’ not AI
• Microsoft finds security flaw in AI chatbots that could expose conversation topics
• Sam Altman predicts AI will create flawless poetry, but no one will care
• The AI boom feels eerily similar to 2000’s dotcom crash – with some important differences
• Wie Europol mit Microsoft, Palantir, Clearview & Co. auf Kuschelkurs geht
• El 11.11 de AliExpress está a punto de comenzar con cupones exclusivos, descuentos de hasta el 80% y chollazos en tecnología

Today’s Pulse on AI dives into Morgan Freeman’s pushback on AI voice cloning and what “consent, compensation, and control” should look like; Google’s awkward AI-generated Bundesliga tickers; an AI support mishap sending gamers to the wrong Obsidian; a study showing Australia leads per-capita AI use; why GPT-4o’s personality can’t be reproduced across training runs; CMU’s EMNLP highlights on agents, retrieval, safety, and steerability; Oracle’s Autonomous AI Lakehouse and what Iceberg means for data teams; major funding across AI-enabled parking, healthcare agents, BCI, and security; a strange ChatGPT privacy leak surfacing prompts in Google Search Console; Nigerian startups localizing AI and data for sales, support, sports, and creators; Kling AI’s upgraded text-to-video with 3D physical realism; and Birlasoft’s nod to “Agentic AI” in enterprise. We close with three takeaways: prioritize consent and provenance in AI media, automate with human guardrails, and win by pairing robust data plumbing with localized design.

Sources:

• ​​Birlasoft Q2 Revenue Rises 3.4% to ₹1,329 Crore; Margin Expands to 16%
• Oddest ChatGPT leaks yet: Cringey chat logs found in Google analytics tool
• Carnegie Mellon University at EMNLP 2025
• Here's one way to cut support ticket volume… send them to another company entirely
• The Week’s 10 Biggest Funding Rounds: A Varied Lineup, Led By Crypto And Parking
• Inside Oracle’s Autonomous AI Lakehouse: Open, intelligent and multicloud by design
• 7 African startups powering sales, sports, support, and settlements
• Kling AI lowers barriers to making AI-powered content creation more accessible
• Why GPT-4o's "personality" can't be recreated
• Is Australia the most AI-obsessed nation in the world?
• Googles KI-Phrasendreschmaschine für Fußball-Ticker
• ‘I enter a room and people say: “God just walked in”’: Morgan Freeman on voicing the divine, meeting Mandela – and his six decades on screen

Max and AI expert Selene Arcaro dive into Google’s File Search tool for Gemini, the rise and risks of “vibe coding,” Pinterest’s shift to fine‑tuned open source models, a practical framework for diagnosing LLM failures, and the MarkItDown utility for creating LLM‑ready Markdown. They unpack Nigeria’s ambitious AI bill, how AI is reshaping jobs, October’s most active investors, Spain’s landmark deepfake sanction, the ecological angle of undersea cable builds, new theory for diffusion sampling with CLD, and whether developers should be forced to use AI tools. Three takeaways: ground your models, pick right‑sized models, and keep learning.

Sources:

• Developers Should Have a Choice of Not Using AI
• Google says project on famous crab-covered island is about cables, not combat
• Wasserstein Convergence of Critically Damped Langevin Diffusions
• Introducing the File Search Tool in Gemini API
• Protección de Datos multa con 2.000 euros al difusor de uno de los desnudos falsos de las chicas de Almendralejo
• 'Vibe coding' named Word of the Year. Developers everywhere faceplant
• How to Diagnose Why Your Language Model Fails
• Active US Investors Kept Busy Cutting Checks In October
• Python MarkItDown: Convert Documents Into LLM-Ready Markdown
• Nigeria’s AI bill proposes mandatory registration and licensing for developers and users
• Pinterest CEO touts open source AI: ‘tremendous performance’ with reduced costs
• KI: Kein Job bleibt wie bisher – Dauerlernen wird zur Norm
• ‘Vibe coding’ beats ‘clanker’ to be Collins dictionary’s word of the year
• La Navidad ha reavivado el fantasma de la reduflación. Y hay una forma fácil de entenderlo: mirar un turrón Suchard de 2020

Today’s Pulse on AI dives into Europe’s “Digitalokratie” debate and AI sandboxes, neuromorphic breakthroughs from USC and BrainChip, and Gemini’s new powers in Google Maps. We share a practical five-step framework to diagnose LLM failures, a look at MarkItDown for LLM-ready documents, and security lessons from AMD’s Zen 5 RNG flaw and Apple’s iOS 26.1 update. Plus: the pitfalls of AI-made ads, human attachment to chatbots, Christian AI ambitions, market jitters, faster diffusion sampling, AI-forward smartphones, and browsers turning into identity managers.

Sources:

• BrainChip Launches AKD1500 Edge AI Co-Processor
• Wasserstein Convergence of Critically Damped Langevin Diffusions
• Tech News Today: AMD’s critical CPU flaw and iOS 26.1 offerings
• From Intel to the infinite, Pat Gelsinger wants Christian AI to change the world
• How to Diagnose Why Your Language Model Fails
• Python MarkItDown: Convert Documents Into LLM-Ready Markdown
• Apple iOS 26.1 update is here: See what’s changing on your iPhone
• Google Maps bakes in Gemini to improve navigation and hands-free use
• AI and you: A match made in binary code
• Digitalokratie: Kann Europa die digitale Freiheit zurückgewinnen?
• Artificial neurons that behave like real brain cells
• Global stock markets fall sharply over AI bubble fears
• Jeff Bezos asegura que hay un tipo de empleado que nunca podrá ser reemplazado por una IA: los inventores
• The best phones of 2025: Expert tested and reviewed