Pulse on AI

NVIDIA discounts Jetson developer kits as real-world edge AI projects—from a self-paddling canoe to underwater fish monitoring and factory humanoids—show what’s possible. Apple’s App Store Awards spotlight AI as an ingredient powering everyday apps like Tiimo, Detail, Strava, StoryGraph, and Be My Eyes. A Streamlit tutorial demonstrates how to productize inventory analytics for operations. The hiring market grapples with AI-fueled application overload, elevating referrals and practical assessments, while ZTE’s CDO outlines a pragmatic path for agentic AI with humans firmly in the loop. A research preprint proposes roundtrip verification to mitigate LLM hallucinations on invertible tasks. German authorities strike at deepfake-driven investment ad networks. The UK’s AI minister pushes faster adoption with guardrails, and NTT’s new Bengaluru data center campus underscores the compute buildout powering AI.

Sources:

• NTT Inaugurates New Data Centre Campus in Bengaluru
• Mitigating hallucinations and omissions in LLMs for invertible problems: An application to hardware logic design automation
• Robots’ Holiday Wishes Come True: NVIDIA Jetson Platform Offers High-Performance Edge AI at Festive Prices
• Apple’s App Store winners show AI is now just a hidden feature
• ZTE CDO Cui Li speaks at Economist Impact's AI Innovation Asia 2025
• UK AI minister Kanishka Narayan: 'We would like to go even faster'
• 👨🏿‍🚀TechCabal Daily – No more Shein-nanigans
• AI finds its way into Apple’s top apps of the year
• Build and Deploy Your First Supply Chain App in 20 Minutes
• From data streams to strategic insight: How Parth Joshi is inventing AI-driven intelligence, user-first product analytics, and real-time monitoring
• Großrazzia: Schlag gegen die Werbe-Industrie des Anlagebetrugs
• El caos que ha generado la IA en la contratación de personal ha revelado un tipo de talento oculto: "desarrolladores invisibles"
• Apple crowned the best apps of 2025 - did your favorite make the list?

Max and AI expert Silvia Rennard break down why Windows 11 adoption is slower than expected despite Windows 10’s end-of-support, Apple’s AI leadership transition as Amar Subramanya steps in amid Siri delays, and Raspberry Pi’s price hikes fueled by AI-driven memory demand. They explore COPE, an open-source chain-of-thought framework for predicting stroke outcomes from clinical notes, and a hands-on k-NN classifier built in Excel. The duo dives into Ridelink’s AI-enabled logistics and embedded finance for SMEs, what to watch at AWS re:Invent 2025, Zig’s move from GitHub to Codeberg over Actions reliability and AI direction, and Huawei/SERES’s AITO M9 overseas rollout with ADS and satellite connectivity. The episode ends with actionable takeaways and a nudge to tap into free AI seminars this month.

Sources:

• Forthcoming machine learning and AI seminars: December 2025 edition
• Amar Subramanya -The Bengaluru Engineer Now Running Apple’s GLOBAL AI Strategy
• COPE: Chain-Of-Thought Prediction Engine for Open-Source Large Language Model Based Stroke Outcome Prediction from Clinical Notes
• Raspberry Pi just got up to $25 more expensive
• Windows 11 still barely pulling ahead of 10 despite end-of-support push
• Overseas Version of AITO M9 Infotainment System Revealed, Complete with Localized Maps, HUAWEI ADS, and Global Apps
• Ridelink targets SME trade with AI logistics and embedded finance
• AWS re:Invent 2025: How to watch and follow along live
• Apple's head of AI resigns after Siri problems
• The Machine Learning “Advent Calendar” Day 2: k-NN Classifier in Excel
• Zig kehrt GitHub den Rücken: Frust über Actions und Microsofts KI-Kurs
• I found the best Cyber Monday Apple Watch deals

Today’s episode compares small and large language models with Microsoft’s latest on‑device SLM, examines privacy and security pitfalls in AI browsers like Atlas, and shows how to orchestrate multiple GitHub Copilot agents using mission control for real throughput gains. We discuss HSBC’s partnership with Mistral for self‑hosted banking AI, the architecture of AI‑native data centers, and new research suggesting brain‑aligned benefits from convolutional networks. We also parse claims about GPT‑5’s scientific problem‑solving, unpack open‑source model definitions, debate Meta’s dominance in an AI context, explore Avandra’s medical imaging data network, highlight edge‑ready IoT anomaly detection with Isolation Forest, and mark ChatGPT’s third anniversary.

Sources:

• Avandra Acquires DatCard Systems and Sorna Corporation
• AI Native Data Centres: Hype or the Making of a Sovereign Compute Era?
• Unsupervised Anomaly Detection for Smart IoT Devices: Performance and Resource Comparison
• ChatGPT turns three today
• How to orchestrate agents using mission control
• HSBC partners with Mistral AI as banking giants spend billions looking for LLM boost
• If Meta Isn’t A Monopoly, Then The Word Doesn’t Mean Anything
• Small changes make some AI systems more brain-like than others
• What are small language models and how do they differ from large ones?
• The Problem with AI Browsers: Security Flaws and the End of Privacy
• When GPT-5 thinks like a scientist
• Open-Source-KI: Was es gibt und wie offen die Modelle wirklich sind
• James Cameron says AI actors are ‘horrifying to me’
• How to take long scrolling screenshots on your iPhone or Android - it's easy

Max and AI expert Clara Mendieta break down how AI is reshaping go-to-market strategies, explain Multi-Token Prediction and why it can make LLMs faster and better at reasoning, discuss Google’s limits on free image and Gemini access, and examine UrSafe’s AI drone safety model for Nigerian schools. They cover Warner’s Suno licensing pivot, SkySparc’s Dubai expansion, TikTok’s algorithmic harm with the Disney “princess diet,” China’s humanoid robot bubble risk, personalization at scale, OpenAI’s Mixpanel data incident, an open-source paraconsistent logic library, Apple overtaking Samsung without strong AI, Germany’s low-risk AI quality standard, and Microsoft Edge’s AI shopping features. Three takeaways: AI amplifies good GTM craft, MTP is a practical new LLM lever with trade-offs, and trust grows when AI is transparent and consent-driven.

Sources:

• SkySparc Strengthens Middle East Presence With New Dubai-based Entity
• Paraconsistent-Lib: an intuitive PAL2v algorithm Python Library
• High demand forces Google to tighten free access to Nano Banana Pro
• La ‘dieta de las princesas Disney’: cuando el algoritmo de TikTok incita a la anorexia y la bulimia
• Apple’s lousy AI didn’t stop it beating Samsung’s smartphone sales for the first time since 2011
• Can AI and drones protect Nigeria’s school children? UrSafe thinks so
• How OpenAI and Google see AI changing go-to-market strategies
• Microsoft adds built-in AI shopping tools to Edge in the U.S.
• Why We’ve Been Optimizing the Wrong Thing in LLMs for Years
• The role of AI in modern marketing: Personalisation at scale
• China says humanoid robot buzz carries bubble risk
• OpenAI meldet Datenklau bei Dienstleister Mixpanel
• Neuer Qualitätsstandard für Niedrigrisiko-KI vorgestellt – Fraunhofer IAIS bringt Expertise zu Prüfung und Technik ein
• Warner ha llegado a un acuerdo con una empresa de IA para que use su catálogo. Los que llevan las de perder son los artistas

Max and AI expert Priya Deshmukh dive into how AI can cut the hidden “noise” in human decisions, from courts to insurance and hiring, and when humans should overrule with decisive context. They cover Big Tech’s rush to hire neuroscientists for efficiency and interpretability, WhatsApp’s ban on third‑party general‑purpose chatbots, cross‑border data sovereignty (OVHcloud vs Ontario), and Mexico’s national supercomputer initiative for climate, satellites, and public‑sector LLMs. Corporate news includes HP’s AI‑driven cost cuts. Research highlights: Harmonic AI’s $120M raise for formal math reasoning with Lean 4 and a study on activation steering showing inverted‑U behavior and the limits of vector metrics. Product updates: Gemini “Projects” workspaces on Android, Speechify’s voice typing and assistant, TierPoint’s VMware Cloud Foundation 9.0 private clouds for AI workloads. Fintech in Africa: AXIAN’s shift from mobile money to full digital banking with AI underwriting. Social impact: how AI can be a lifeline for blue‑collar workers by parsing nontraditional résumés, verifying credentials, and prioritizing skills over polish. Three takeaways: reduce noise before adding complexity, use AI as a consistency engine with human overrides, and remember platform and product design choices determine who benefits.

Sources:

• TierPoint Launches Next-Generation VMware Platform for Private Clouds
• 7 Must-Read Books on AI in 2025
• Steering Latent Traits, Not Learned Facts: An Empirical Study of Activation Control Limits
• Gemini’s upcoming Projects feature mirrors ChatGPT workspaces
• HP to sack up to six thousand staff under AI adoption plan, fresh round of cost-cutting
• White-Collar Workers Fear AI, But For Blue-Collar Workers, It Can Be A Savior
• Harmonic AI raises $120M at $1.45B valuation to advance mathematical reasoning
• AXIAN rebrands fintech arm in digital banking push across Africa
• Speechify adds voice typing and voice assistant to its Chrome extension
• Meta removes rival chatbots from WhatsApp
• Everyday Decisions are Noisier Than You Think — Here’s How AI Can Help Fix That
• Mystery foot belongs to ancient human relative: scientists
• Mittwoch: Internationaler Streit um Cloud-Daten, WhatsApp bald ohne externe KI
• Los grandes de la IA se están disputando a los neurocientíficos como si fueran estrellas de fútbol

Max and AI expert Mira Solberg unpack Ilya Sutskever’s claim that the “age of scaling” is ending and why research breakthroughs—not just more chips—may drive the next leap. They examine HSBC’s estimate that OpenAI needs $207B of new financing by 2030, with ripple effects for Oracle, Microsoft, Amazon, Nvidia, AMD, and SoftBank. They break down Meta’s interest in Google TPUs and Nvidia’s response, discuss a New York court order requiring OpenAI to disclose internal legal communications related to deleted book datasets, and cover WhatsApp’s new policy shutting out general-purpose AI chatbots like Copilot. The episode explores Huawei’s Mate 80 series and on-device AI imaging, Kovant’s agentic SLM swarms for enterprises, and the booming market for “screen-free” AI toys including Bondu, Roybi, and Stickerbox. They explain the viral trend of making Stranger Things-style portraits with Google’s Nano Banana via Gemini, share a practical workflow using AI in Cursor to tame LaTeX documents, and dissect the EPA’s plan to prioritize data-center-related chemicals amid concerns about PFAS in immersion cooling. Finally, they look at “AI slop” as 2025’s word of the year and what it means for trust and quality online. Key takeaways: expect research-driven progress, infrastructure choices and policies will shape winners, and users should demand privacy and provenance—especially for kids’ tech.

Sources:

• Ilya Sutskever Changes His Mind About How to Build Superintelligence
• Copilot exits WhatsApp on January 15 citing policy shift
• HSBC spies $207B crater in OpenAI's expansion goals
• The Toy Startups Getting Funded This Year: Talking Dinosaurs, Sticker Cubes And Tin Can Phones
• Kovant wants to become the nerve center of enterprise agentic AI operations after pre-seed funding
• 👨🏿‍🚀TechCabal Daily – Uber goes Electric
• Huawei rolls out Mate 80 series featuring new Kirin 9030 chip and ultra-bright display
• NYC judge: OpenAI must turn over communication with lawyers about deleted databases
• AI slop is Macquarie’s 2025 Word of the Year. I applaud the choice – but was bored by the shortlist
• How to Create Professional Articles with LaTeX in Cursor
• Meta vor milliardenschwerem Kauf von Googles KI-Chips? Nvidia-Aktie sackt ab
• Ilya Sutskever breaks silence on AI's future
• The Trump Administration’s Data Center Push Could Open the Door for New Forever Chemicals
• Cómo convertir cualquier foto tuya en un personaje de Stranger Things usando Nano Banana

Max and guest expert Celeste Morrell unpack a packed day in AI: Disney+ and Hulu’s ad-driven bundle and the algorithms behind streaming economics; senators calling for investigations into Meta’s alleged scam-ad profits and Meta’s denial; OpenAI’s “Cameo” trademark snag and the ethics of consent in deepfakes; “AI slop” as Macquarie Dictionary’s word of the year; a push for better AI imagery on book covers; HKUST’s humanoid layup demo and the imitation-learning advances behind it; Llamazip’s lossless compression via LLaMA and what it implies for training-data provenance; AWS’s 1.3-gigawatt government-grade AI datacenters; H2O.ai’s leadership move amid sovereign AI momentum; using LLMs as judges to evaluate other models; AI-generated music scaling on Spotify; India’s AI-healthcare startups delivering clinical impact; EU tech policy turbulence; and a new physics benchmark showing top models still struggle with original research. Three takeaways: design for transparency, calibrate hype with repeatability, and keep governance practical.

Sources:

• H2O.ai Appoints Jason Finney as President and Chief Revenue Officer
• Better images of AI on book covers
• Top 10 Indian Startups Powering Healthcare with AI
• Llamazip: Leveraging LLaMA for Lossless Text Compression and Training Dataset Detection
• Meta accused of earning 10% of revenue from known scams
• AWS to build 1.3 gigawatts of government-grade supercomputing power for Uncle Sam
• Unitree G1 Nails a Perfect Layup: HKUST Showcases the World’s First Real-World Basketball Robot Demo
• OpenAI learned the hard way that Cameo trademarked the word ‘cameo’
• Gemini 3 Pro and GPT-5 still fail at complex physics tasks designed for real scientific research
• LLM-as-a-Judge: What It Is, Why It Works, and How to Use It to Evaluate AI Models
• KI-Musik: Wenn Algorithmen den Hit des Sommers komponieren
• Macquarie Dictionary announces ‘AI slop’ as its word of the year, beating out Ozempic face
• Europe Is Bending the Knee to the US on Tech Policy
• Disney+ and Hulu bundled for $5 a month might be my favorite Black Friday streaming deal

Max and Soraya break down five hard-won ways to stop your AI strategy from going bust, then tackle bubble fears and Google’s plan to 1000x compute. They dive into Amazon’s multi-agent security system (ATA), why the next AI wave belongs to infrastructure and compliance, and Apple’s stability-first iOS reset for better on-device AI. They compare Google’s Gemini 3 to ChatGPT, explore Xiaomi’s open-sourced model unifying robots and autonomous driving, and explain how explainable AI could make self-driving safer. Plus: entropy-guided hybrid modeling, AI for architecture diagrams, and a lesson in transparent AI from the Redford family.

Sources:

• ‘Holy S***… I’m Not Going Back to ChatGPT,’ Says Marc Benioff After Using Gemini 3
• BITS for GAPS: Bayesian Information-Theoretic Sampling for hierarchical GAussian Process Surrogates
• Apple prepares a system-wide code cleanup in iOS 27 to pave the way for AI
• Why AI’s Next Phase Belongs To Infrastructure
• Xiaomi Releases and Fully Open-Sources MiMo-Embodied, the First Model to Bridge Autonomous Driving and Embodied Intelligence
• Safer Autonomous Vehicles Means Asking Them the Right Questions
• Google plans a 1000x jump in AI compute over the next five years
• Zwei Folgen software-architektur.tv: KI in Dokumentation und Architektur
• ‘Extra challenging during a difficult time’: Robert Redford’s daughter criticises AI tributes to the late actor
• Amazon Is Using Specialized AI Agents for Deep Bug Hunting
• Invertimos en IA como si ya hubiera triunfado. Es exactamente lo que hicimos en 1999
• 5 ways to prevent your AI strategy from going bust

Today’s episode dives into AI-edited real estate photos on platforms like Idealista and why disclosures matter; how automakers fall into the optimization trap instead of reimagining cars with generative AI; OpenAI’s launch of group chats in ChatGPT; Deepwatch’s new Bengaluru engineering hub for MDR; Fermat’s open-source RL environment for automated math discovery and EvoAbstract’s learned “interestingness”; Grok’s sycophancy toward Elon Musk and trust implications; the federal preemption push for AI laws; a French consortium’s model-reduction and data-assimilation research for faster simulations; Cisco’s campaign against legacy infrastructure risks in an AI era; AnyLanguageModel’s unified Swift API bridging local and cloud LLMs on Apple platforms; Nvidia’s earnings beating doubts amid longer-term questions; Amazon Bedrock Guardrails’ code-domain protections; and Microsoft’s “AI-enabled Cloud PC” plus expanded hybrid AVD. Three takeaways: trust is the new UX, reimagination beats incrementalism, and edge-cloud convergence is defining where AI runs and how safely.

Sources:

• Why Deepwatch Chose Bengaluru to Capture MDR Market
• Trump revives unpopular Ted Cruz plan to punish states that impose AI laws
• Learning Interestingness in Automated Mathematical Theory Formation
• Amazon Bedrock Guardrails expands support for code domain
• OpenAI turns ChatGPT into a social network with global group chats
• Microsoft reveals new cloudy AI PC that’s not a Copilot+ PC
• Introducing AnyLanguageModel: One API for Local and Remote LLMs on Apple Platforms
• Grok says Elon Musk is better than basically everyone, except Shohei Ohtani
• Generative AI Will Redesign Cars, But Not the Way Automakers Think
• Nvidia vor Quartalsbericht: Warum Anleger genau hinsehen sollten
• Accélérer la simulation numérique dans des secteurs clés comme l’aéronautique et la défense
• Nvidia earnings: Wall Street sighs with relief after AI wave doesn’t crash
• With the Rise of AI, Cisco Sounds an Urgent Alarm About the Risks of Aging Tech
• Idealista se está llenando de imágenes de casas a la venta hechas con IA. Y cada vez es más difícil identificarlas

Yann LeCun departs Meta to found an AMI-focused startup with Meta as a partner; Nvidia posts massive revenue and margins while investors debate GPU depreciation and circular deals; Hugging Face’s CEO argues we’re in an LLM bubble but not an AI bubble; GitHub Copilot boosts success and speed by shrinking toolsets and routing via embeddings; Anthropic’s Claude Code on AWS Bedrock: direct IdP auth, a dedicated account, and OpenTelemetry monitoring; the EU proposes delaying enforcement for high‑risk AI until standards are finalized; Apple’s iPhone Air designer exits to an AI startup; JD.com launches an AI‑powered review platform integrated with delivery; an AI‑enabled grill shows utility vs gimmick; OpenAI board governance in the spotlight; AI image hoaxes mislead travelers; the series Plur1bus stirs reflection on collective intelligence; and a careful discussion of AI’s impact on sexuality, emphasizing information quality, bias, and consent.

Sources:

• “We’re in an LLM bubble,” Hugging Face CEO says—but not an AI one
• Claude Code deployment patterns and best practices with Amazon Bedrock
• iPhone Air designer Abidur Chowdhury just left Apple for an AI startup
• La UE concede a las tecnológicas 16 meses más para que su IA de “alto riesgo” cumpla la normativa
• How we’re making GitHub Copilot smarter with fewer tools
• It's a good time to be the arms dealer for the AI boom
• JD.com Enters Local Services Fray with "JD Review" Platform
• The Epstein files have claimed their first OpenAI victim: board member Larry Summers
• Vince Gilligan’s sci-fi series Plur1bus taps into our greatest fears about AI
• Developing Human Sexuality in the Age of AI
• Metas Chefwissenschaftler geht: LeCun gründet ein Start-up
• Buckingham Palace Christmas market: why tourists flocked there – and found just locked gates and big puddles
• The Zelos-450 Pellet Grill Has Features Missing on Grills Triple Its Price
• For the first time, the Apple Watch Series 11 is priced to recommend without hesitation