AI NEWS YOU MISSED ?#58

Top 5 stories of the?week

1?? Alibaba tops?DeepSeek?

2?? GPT-4.5?#1

3?? Grok 3 #1?too

4?? Superintelligence: Nuclear Arms Race of?AI

5?? Try Sesame: Can You Tell It’s Not a?Person?

VIDEO SUMMARY?

Details on Top 5 stories of the?week

1?? Alibaba tops?DeepSeek

Qwen’s QwQ-32B Model Redefines Open-Source AI Leadership

• Efficient Performance:?Despite having only 32 billion parameters, QwQ-32B matches the capabilities of larger models like DeepSeek-R1, which has 671 billion parameters, highlighting advancements in AI efficiency.
• Cost-Effective Deployment:?QwQ-32B’s smaller size allows for reduced computational requirements, leading to lower operational costs and making high-performance AI more accessible.
• Strategic Implications:?The success of QwQ-32B underscores a shift towards developing smaller, more efficient models that deliver competitive performance, challenging the notion that larger models are inherently superior.

Access the model on Hugging Face

Alibaba stock price up 65% in 2025

2?? GPT-4.5?#1

GPT-4.5 Beats Competitors in Blind Tests

Despite early skepticism, users evaluating the AI’s responses have noted a distinct vibe and strikingly human nature in its interactions, as the GPT-4.5 outranks rivals.

“It’s a lemon”?—?OpenAI’s largest AI model ever arrives to mixed reviews

A critical analysis of OpenAI’s latest model, arguing that GPT-4.5 offers only marginal improvements over previous versions at a much higher cost.

3?? Grok 3 #1?too

Grok-3 Takes the Lead in AI Performance Rankings

xAI’s latest Grok-3 model has topped the Arena leaderboard, tying for #1 across multiple categories?—?including Hard Prompts, Coding, Math, and Creative Writing?—?marking a major milestone for Musk’s AI venture.

DeepSeek & Grok Fever Over? ??

DeepSeek and Grok have rapidly lost momentum in the AI app rankings amid data privacy concerns and controversy, signaling ongoing uncertainty about which AI players will ultimately dominate.

4?? Superintelligence: Nuclear Arms Race of?AI

READ

A new report by Dan Hendrycks, Eric Schmidt, and Alexandr Wang warns that superintelligence could destabilize global security and outlines strategies to mitigate its risks.

• Mutual Assured AI Malfunction (MAIM):?A deterrence framework where states may sabotage rival AI projects to prevent unilateral dominance.
• AI Nonproliferation:?Stricter controls on AI chips, model access, and cybersecurity to prevent rogue actors from weaponizing AI.
• Geopolitical Competitiveness:?Nations must secure AI leadership through domestic chip manufacturing, military AI integration, and legal frameworks for AI governance.

5?? Try Sesame: Can You Tell It’s Not a?Person?

Try here

Sesame’s Conversational Speech Model aims to bridge the uncanny valley in voice assistants by incorporating emotional intelligence, contextual awareness, and conversational dynamics, although significant progress is still needed to fully replicate human prosody and conversational interaction.

INSEAD Programs

AI for Business

The INSEAD AI for Business Programme equips participants to navigate AI’s transformative impact by offering a comprehensive understanding of AI?—?from its technical foundations and strategic opportunities to organizational challenges?—?enabling managers to effectively leverage AI technologies and lead confidently in the digital era.

Transforming Your Business with AI

Gain a deep understanding of how AI is deployed in business, what it can and cannot do, to make the most of these emerging technologies.

Upcoming Live Events?—?INSEAD?related

The AI Adoption Challenge: What’s Holding Enterprises Back??—?Thursday, March 13

Join Frederic Brunier and Emmanuel Viale from Accenture, along with Lily Fang from INSEAD, as they discuss the key barriers enterprises face in adopting AI at scale, and explore actionable strategies to accelerate implementation and unlock AI’s full business value.

Transforming Banking with AI & Prompt Engineering?—?Thursday March 27

In this 60-minute Tech Talk on March 27, 2025, at 10:00 am CET, Royce Teo (DBS), Gianluca Malagoli (Accenture), and Loic Sadoulet (INSEAD) will explore how AI and prompt engineering are transforming the banking industry, from customer experience to fraud prevention.

More AI stories of the?week

Suggested by INSEAD-ers. From the 3,000+ person global INSEAD AI private business leader community, regularly sharing AI-related articles, vids, etc, of interest. (alum-led)

China and US need to cooperate on AI or risk ‘opening Pandora’s box’, ambassador warns

Xie Feng warns of the danger of letting the technology develop unchecked and says the countries should cooperate on global governance

Turing Award winners warn over unsafe deployment of AI models

Two pioneers, Sutton and Barto, authors of the foundational reinforcement learning book, have won the $1mn prize from the Association for Computing Machinery.

READ the book

10 Surprising AI Public Views

A majority sees AI positively impacting society, yet widespread concerns about privacy, bias, job displacement, and low trust in tech companies’ ethics persist, particularly highlighting high confidence in healthcare AI.

Content Moderation and Governance: Principles, Misconceptions, and Challenges

A detailed exploration into the challenges and complexities of content moderation, emphasizing the critical importance of fair processes, transparency, and independent oversight, especially during crises, to safeguard freedom of expression while ensuring online safety.

LinkedIn summary

INSEADer Lutz Finger on AI: Competitive Edge Is User Experience, Not Model Size?

The real competitive advantage in AI isn’t bigger models like Musk’s $12 billion Grok?—?which offers minimal improvements over rivals?—?but deeply integrating AI into users’ workflows through seamless, user-focused design.

Boards: Is Your Talent Strategy Keeping Pace with AI?

Boards must urgently prioritize AI readiness by investing in leadership, employee training, targeted hiring, and performance management to ensure organizational competitiveness.

Where Does AI Get Its Knowledge From??—?INSEADer Benjamin Latli LLMs are software programs trained to generate human-like text based on patterns learned from vast amounts of data, stored as files and loaded into computers, gaining public trust through their apparent knowledge despite concerns about privacy, bias, and the ethical management by tech firms.

How to make Chinese AI work for the US?—?with Jeffrey Ding?(Azeem Azhar)

• AI diffusion, not just innovation, is key?—?Jeffrey Ding emphasizes that widespread adoption and integration of AI into industries matter more for economic and geopolitical impact than just leading in AI innovation.
• China’s AI growth is accelerating but faces challenges?—?DeepSeek has gained rapid adoption in China, but questions remain about its actual productivity benefits versus state-driven hype, and diffusion in enterprise use is still limited.
• U.S. containment strategies may be ineffective?—?Current U.S. export controls on AI chips and technology haven’t significantly slowed China’s AI progress, and fostering domestic AI talent and infrastructure may be a better strategy.

AI companies race to use ‘distillation’ to produce cheaper models

DeepSeek used technique to create smaller powerful models based on the technology of competitors such as Meta

AAAI 2025 Presidential Panel Report?—?the Future of Research

The AAAI 2025 Presidential Panel report highlights the rapid transformation of AI research, emphasizing the need for interdisciplinary collaboration, rigorous evaluation, improved factuality and trustworthiness, ethical frameworks, and international cooperation to safely and responsibly harness AI’s expanding capabilities

Microsoft’s AI Muse for Game Design Microsoft introduces an AI model specifically designed to assist in game development and ideation.

Microsoft Muse: Their first generative AI model designed for gameplay ideation Microsoft introduces an AI model specifically designed to assist in game development and ideation.

NVIDIA sales to China +Taiwan + Singapore comparable to US sales.

Everything else is relatively small

News on purported smuggling from Singapore to China

Apple’s M3 Studio Ultra: 512GB Unified RAM for Your Home AI Lab

A humorous guide to negotiating the purchase of Apple’s powerful new M3 Studio Ultra with 512GB Unified RAM, emphasizing its game-changing ability to run massive AI models privately at home.

Wharton AI Lab: Prompt Engineering Findings

Wharton’s Generative AI Lab’s first Prompt Engineering Report reveals that common prompting tricks, like adding “please,” do not consistently improve outcomes, and emphasizes the importance of selecting benchmarks tailored to specific use cases.

AI at the Oscars – Predictions & Forecasts?by Grok AI-powered predictions for the 2025 Academy Awards.

World and Human Action Models towards gameplay ideation?

A new generative AI model, the World and Human Action Model (WHAM), aligns with creative needs in game development by generating consistent and diverse gameplay sequences, allowing user modifications to persist, thereby enhancing iterative ideation processes.

AI vs. Human Worldview Benchmark

A test comparing AI models to human performance on factual knowledge, showing that humans often do worse than chance.

Probabl: Why Accuracy Metrics Often Don’t Matter in Business?—?(CEO is INSEADer Yann Lechelle)

Standard AI accuracy metrics like F1 score are often overrated in real-world applications, as businesses care more about tangible outcomes like stability, cost reduction, and profit?—?highlighting the need for custom metrics aligned with business impact.

Introduction to AI Safety, Ethics, and Society?By?Dan Hendrycks

Provides a comprehensive guide to AI risks, including malicious use, system failures, and regulatory challenges, emphasizing the need for cross-disciplinary collaboration to ensure safe and responsible AI development.

Why I’m Concerned with Open AI (and what no one seems to be talking about)

Stephen Klein raises concerns about OpenAI’s financial losses, lack of a sustainable business model, ethical controversies, and competitive weaknesses, questioning whether its hype-driven narrative is masking deeper structural issues and an unsustainable AI bubble.

DeepSeek FAQ?—?AI’s Algorithmic Innovations A breakdown of how DeepSeek optimizes AI performance on lower-spec chips, bypassing U.S. export restrictions.

Day 6: One More Thing, DeepSeek-V3/R1 Inference System Overview A technical overview of DeepSeek’s new AI inference system and its potential to disrupt existing AI infrastructure.

Profiling Data in DeepSeek Infra

DeepSeek theoretical efficiency

Now I get why Microsoft scrapped projects for the new AI datacenter

Microsoft’s Phi-4-multimodal-instruct is a compact yet powerful AI model with a 128K token memory, trilingual inputs, and multimodal capabilities, designed for on-device use, enabling speech translation, OCR, multi-image analysis, and reasoning?—?while being open-sourced under MIT license, signaling a major shift in AI accessibility and control.

AI agents could birth the first one-person unicorn?—?but at what societal cost?

AI agents could enable solo entrepreneurs to build billion-dollar companies by automating tasks, scaling operations, and collaborating autonomously, raising questions about the future of employment and societal impact as AI-driven businesses challenge traditional workforce structures.

Anthropic Faces Backlash Over Data Collection in AI Alignment

Anthropic AI is under fire for allegedly collecting user data via its Computer Use API without informed consent, later updating its website to appear transparent, raising concerns about ethical AI alignment and user privacy.

D.O.G.E. going after consulting contracts

Two dozen arrested in international swoop for links to AI-made child sex abuse images

Two dozen people were arrested in an international crackdown targeting a criminal network distributing AI-generated images of child sexual abuse, marking one of the first major law enforcement operations against such material.

The Anthropic Economic Index A deep dive into how Anthropic’s AI models are being used, with insights into their user demographics and application trends.

Evaluating Intelligence via Trial and Error

The Survival Game framework measures intelligence by assessing how efficiently an entity?—?whether biological or artificial?—?finds solutions through trial and error, revealing that while AI can autonomously solve simple tasks, it remains far from achieving human-like adaptability in complex domains due to exponential scaling challenges.

Start-up says its ‘woolly mice’ bring mammoth revival closer

Scientists warn big obstacles, including ethical questions, remain in efforts to restore extinct species

Data Science Agent in Colab: The future of data analysis with Gemini

Google Colab’s new Data Science Agent, powered by Gemini, automates data analysis by generating complete, executable Jupyter notebooks from natural language prompts, streamlining workflows and saving researchers time on setup and coding.

Microsoft Unveils MAGMA-8B: A Game-Changing Vision-Language Model

Microsoft’s MAGMA-8B is a powerful multimodal AI that integrates vision, language, and action, enabling applications in robotics, web automation, and beyond, using innovative Set-of-Mark and Trace-of-Mark prompts for action grounding and planning.

Build Apps with Windsurf’s AI Coding Agents

This course teaches developers how to leverage Windsurf’s AI coding agents to streamline workflows, debug and refactor code, update large codebases, and build applications efficiently by integrating AI-assisted search, context awareness, and multi-file code generation.

AI-Powered UI Design: Cursor + Claude 3.7 (in Chinese)

This tutorial demonstrates how Cursor and Claude 3.7 enable users to effortlessly generate full product prototypes and UI designs, making AI-driven design more accessible to non-experts.

AI Videos of the?Week

Check out the week’s top AI Videos on the AI-Essential Vids section of the INSEAD AI YouTube Channel which is regularly updated.

The Government Knows AGI is Coming | The Ezra Klein Show

1. AGI Imminent: Experts and government insiders warn that AGI, or AI systems capable of surpassing humans in most cognitive tasks, could arrive within 2–3 years, possibly under the Trump administration.
2. National Security at Stake: The U.S. sees a critical need to achieve AGI ahead of China to avoid potential military, economic, and intelligence disadvantages.
3. Policy Uncertainty: Despite recognizing AGI’s transformative potential, there remains uncertainty about how labor markets, security, and governance structures should prepare for its impact.
4. AI Safety vs. Acceleration: The Biden administration emphasized caution and safety, whereas the incoming Trump administration, supported by figures like Marc Andreessen and J.D. Vance, prioritizes rapid development and innovation, even at higher risk.
5. Government and AI: Current federal structures are seen as too slow to adopt AI effectively, prompting concerns that significant reforms or restructuring may be necessary to harness AI’s capabilities fully.
6. Geopolitical Stakes: Maintaining U.S. leadership in AI development is seen as crucial for national security and global influence, particularly against competition from China, prompting export controls and other strategic measures.

Jonathan Ross, Founder & CEO @ Groq: NVIDIA vs Groq?—?The Future of Training vs Inference | E1260

1. Groq’s Vision: Inference Over Training?—?Jonathan Ross argues that AI inference will be far more critical than training, positioning Groq to dominate low-cost, high-speed inference while NVIDIA retains its grip on training.
2. The Future of AI Hardware?—?Groq’s Language Processing Units (LPUs) challenge traditional GPU architectures by improving efficiency and removing memory bottlenecks, enabling faster and cheaper AI inference at scale.
3. Scaling Challenges and Industry Mismatches?—?Rapid AI growth is constrained by chip supply, power infrastructure, and data center capacity, with overinvestment in some areas and critical shortages in others leading to inefficiencies.
4. The AI Investment Bubble and Market Shifts?—?Ross predicts massive cash burn in the AI sector, but believes the long-term economic returns will outweigh losses, shifting value toward efficient infrastructure providers.
5. China’s AI and Europe’s Position?—?While China is advancing rapidly with DeepSeek and government-backed AI initiatives, Europe risks falling behind unless it fosters a stronger entrepreneurial ecosystem and risk-taking culture.

For a broad selection of AI videos,?check out the INSEAD AI YouTube Channel?(regularly updated)

Past editions?—?AI NEWS YOU?MISSED

View list

SUBSCRIBE TO GET UPDATES IN YOUR?MAILBOX

Follow Robert Maciejko on?LinkedIn?or?X (Twitter)