AI NEWS YOU MISSED ?#53

AI NEWS YOU MISSED ?#53


Top 5 stories since Dec.?21

1?? OpenAI o3 Implications?—?Hints of?AGI?

2?? China’s new AI model breakthrough

3?? Freedom of Speech or War on?Truth?

4?? NVIDIA: AI, Robotics, and Gaming Redefined

5?? Anthropic raising $2B at $60B Valuation


Listen to a NotebookLM Summary

Details of Top 5?stories

1?? OpenAI o3 Implications?—?Hints of?AGI?

Sam Altman expresses confidence in building Artificial General Intelligence

In a recent blog post, OpenAI CEO Sam Altman reflects on the company’s evolution, expressing confidence in building Artificial General Intelligence (AGI) and anticipating the integration of AI agents into the workforce by 2025, while acknowledging past challenges and emphasizing the importance of safety and alignment in AI development.


Microsoft and OpenAI have a financial definition of AGI

OpenAI and Microsoft define AGI as generating $100 billion in profits, far from traditional technical definitions, suggesting Microsoft will retain access to OpenAI’s models for years as the startup remains unprofitable.


Navigating the Policy Challenges of Imminent AGI?—?OpenAI’s Yo Shavit

As AGI becomes plausible, key policy challenges include decentralization of ASI, taxing AI labor through corporate rates, enforcing laws around compute ownership and access, and prioritizing technical alignment to ensure secure and trustworthy AGI development.


Skeptical view from Gary Marcus


2?? China’s new AI model breakthrough

READ MORE

  • DeepSeek AI Breakthrough: China’s DeepSeek, a free open-source AI model, outperforms Meta’s Llama 3.1 and OpenAI’s GPT-4 on key benchmarks, trained for just $5.5 million using lower-performance Nvidia H800 GPUs.
  • Shrinking Barriers in AI: DeepSeek demonstrates that competitive AI models can now be developed at minimal cost with existing outputs, challenging the dominance of major players like OpenAI and Meta.
  • Geopolitical Impact: This advancement highlights China’s growing capabilities in AI development, raising concerns about the competitive edge and implications for U.S.-led AI dominance.


DeepSeek Launch Tweet


DeepSeek-V3: Redefining AI with R1 Distillation and Hardware Insights

DeepSeek-V3 excels in benchmarks, leverages R1 distillation for efficiency, and introduces groundbreaking suggestions for chip manufacturers to optimize AI training and deployment.


China’s Open Source AI: Big Questions, Few Answers

China’s release of an open-source ChatGPT 4o-level model raises critical questions about economic competition, global acceptance of CCP-approved AI, the blocking of U.S. models, and the societal implications of accessible AGI with Einstein-level capabilities.



DeepSeek v3: A Cost-Effective Alternative to GPT-4o and Claude 3.5 Sonnet

DeepSeek v3 emerges as a strong contender for AI applications, offering comparable performance to GPT-4o and Claude 3.5 Sonnet in reasoning, math, and cost efficiency, while excelling in affordability and suitability for AI app development.


Try DeepSeek


3?? Freedom of Speech or War on?Truth?

MUST WATCH?—?Thoughts?

ANTI:?WIRE Take: No Fact-Checking and More Hate Speech: Meta Goes MAGA

  • Meta’s Shift in Moderation Policies: Mark Zuckerberg announced a rollback on fact-checking and hate speech moderation, framing it as a push for “free expression,” aligning with right-wing political rhetoric.
  • Political Context and Moves: Meta has embraced a pro-Trump stance, with leadership changes, a shift of moderation teams to Texas, and policy changes favoring conservative values, including leniency on controversial speech about gender and identity.
  • AI Moderation Rollback: Despite advancements in AI-based hate speech removal, Meta plans to reduce proactive content moderation, focusing only on “high severity” violations, while permitting previously restricted hateful comments.


PRO:?Fox News Opinion: Meta’s Zuckerberg makes a free speech move that could be truly transformational

  • Policy Shift at Meta: Mark Zuckerberg announced Meta’s adoption of X’s free speech standards, ending its fact-checking program, and replacing it with a “community notes” system to focus on removing criminal and fraudulent content.
  • Political Influence: This move coincides with a political shift favoring free speech under the Trump administration, signaling a strategic alignment to avoid regulatory scrutiny.
  • Global Implications: The announcement is seen as a potential turning point in reversing censorship trends, though critics remain skeptical of Meta’s true commitment to free speech.


Meta’s Game: Power, Not Speech?—?Opinion

Meta’s strategy is not about free speech or moderation but about Mark Zuckerberg securing power by offering Meta as a global tool for influence in a loyalty-driven, post-rule-based world.


Misinformation (relevant to the first 3 points here) is a key concern of AI Researchers



What If Zuckerberg’s AI Becomes the Standard?

Mark Zuckerberg’s ambition to make Meta’s AI the global standard raises concerns about privacy, political influence, erosion of choice, and stifled competition, emphasizing the need for transparency, user control, and global AI safeguards to prevent a dystopian future.


Meta’s New Policy: Is Zuckerberg Aligning with Trump?

Meta, under Mark Zuckerberg, is replacing its fact-checking program with a “Community Notes” system, sparking debates about free speech, misinformation, and political influence, with critics suggesting the move aligns with the new Trump administration’s priorities.



4?? NVIDIA: AI, Robotics, and Gaming Redefined

Nvdia’s CES 2025 Event: Everything Revealed in 12 Minutes

  • Next-Gen GPUs and AI Chips: Nvidia unveiled the GeForce RTX 50 Series Blackwell architecture, boasting 92 billion transistors and 4,000 TFLOPS of AI power, enabling advanced AI-graphics workloads. The RTX 5070 offers 4090-level performance at a lower price, with models for laptops also introduced.
  • Foundation Models and Robotics: Nvidia launched the Llama Nemotron and Cosmos World Foundation Models, focusing on physical dynamics, multimodality, and robotics training. The Hyperion 9 platform supports autonomous vehicles with the powerful Thor processor, 20x the capability of its predecessor.
  • Project Digits AI Supercomputer: Featuring the top-secret GB 110 chip, this AI supercomputer integrates Nvidia’s full AI stack and supports DGX Cloud, promising significant advancements in AI research and deployment.


READ: NVIDIA Project DIGITS

A Grace Blackwell AI Supercomputer on your desk


5?? Anthropic reportedly in talks to raise $2B at $60B valuation

READ

Anthropic, an OpenAI rival, is reportedly raising $2 billion in funding at a $60 billion valuation, led by Lightspeed Venture Partners, potentially making it one of the most valuable U.S. startups.


Upcoming live?events

Applied AI in Vertical Apps?—?Friday Jan. 10, 2024 (RSVP LINK)

This seminar by Michael Olenick highlights how LLMs are becoming commoditized, emphasizing the future lies in building specialized, value-added vertical applications to enable strategic thinking, with live demos and Q&A.


AI Private Briefing: Hype, Hope, Headaches, Homework?—?Spaces Limited?—?SIGN UP NOW (click here)?—?Thursday, Jan. 16

  • Exclusive AI Insights: Join for a one-hour private briefing exploring generative AI, business strategies, and governance challenges in 2025.
  • Actionable Takeaways: Discover key predictions, practical steps to prepare your organization, and personal resources to boost your AI expertise, tailored for CEOs, Boards, and senior management.
  • Free, Live-Only Event: Don’t miss this unique opportunity to stay ahead in the AI era?—?limited seats available, so register now to secure your spot!


Are you ready for where AI will take you?—?Jan 19, 2024

Tim Gordon shares insights on Generative AI’s current state, its strategic and economic impact, practical rules for adoption, and potential risks, while offering a broader perspective on societal and policy implications in this 1-hour session.


Strategic Thinking: The AI Event Horizon?—?Feb. 5

Professor Jason Davis explores strategic approaches to harnessing opportunities and mitigating risks at the forefront of AI innovation.


INSEAD AI Happy Hour?—?San Francisco Feb. 7

For those attending the?INSEAD?Americas Conference in San Francisco, we’ve added an INSEAD AI Happy Hour in on Friday, February 7, at 7:30 PM. Over 40 are signed up.

Message me for the RSVP link or check INSEAD AI, SF, Florida or NAA private groups.


More AI stories of the?week

Suggested by INSEAD-ers. From the 1,500+ person global INSEAD AI private business leader community, regularly sharing AI-related articles, vids, etc, of interest. (alum-led)

Kicking Off 2025: Share Your Insights and Ventures?—?from INSEAD SF Hub Director Vic Woo

Start the new year by sharing your thoughts on key questions for business leaders in the AI era and showcase your INSEAD-founded startup through the provided links.


The INSEAD Song (click for Version 1)

The celebratory anthem for Suno AI captures the vibrant energy of the INSEAD alumni community, blending humor, innovation, and collaboration as they lead AI’s golden age, break tech barriers, and inspire a global spirit of discovery and camaraderie.

Version 2


The Golden Opportunity for American AI?—?Microsoft’s Brad Smith

American AI has a golden opportunity to lead globally through advancements in infrastructure, workforce skilling, and international collaboration, ensuring economic growth, innovation, and strategic global influence.


Microsoft expects to spend $80 billion on AI-enabled data centers in fiscal 2025

Microsoft plans to invest $80 billion in fiscal 2025 to build AI-enabled data centers, with over half of the spending focused in the U.S., underscoring its commitment to leading the global AI race through infrastructure, partnerships, and innovation.


AI is Underhyped?—?ex-Google CEO Eric Schmidt

Within the next year, large-context LLM agents with short-term memory and text-to-action capabilities will revolutionize industries by enabling deep understanding, rapid iterative learning, and seamless digital execution, far surpassing the societal impact of social media.


a16z Best AI Apps Now


INSEADer with #1 AI Product- Flowdrafter: Write Faster by Focusing on Drafts?—?Matt Collins

Flowdrafter, a free productivity tool ranked #1 on Product Hunt, helps users write more efficiently by disabling edits and hiding previous text, encouraging quick first drafts that can be refined later.


Matt featured by Menlo Ventures Deedy Das on X


Matt’s Product features in Neuron Daily


Yet another INSEADer AI Product?—?bebooq by Cesare Rotundo


In a Bubble?


OpenAI pro unprofitable


Sam Altman Addresses Family Lawsuit

Sam Altman and his family released a statement expressing their concern for his sister’s well-being, detailing their ongoing support efforts, refuting her allegations, and seeking understanding amidst the legal action she has taken against him.


OpenAI’s Value Capture Challenge in the GenAI Ecosystem

While OpenAI has created a powerful value generation machine, its long-term value capture strategy remains unclear as infrastructure providers and established user interfaces dominate the GenAI ecosystem, leaving LLMs to navigate a challenging middle layer.


Google.org Launches Generative AI Accelerator

Google.org is accepting applications until February 10, 2025, for its Generative AI Accelerator program, offering nonprofits, civic entities, academic institutions, and social enterprises a share of $30M, technical support, and Google Cloud credits to develop socially impactful AI solutions.


AI Survey: Never Mind the Skeptics?—?Bain

Bain’s quarterly survey reveals that despite growing skepticism, executives remain confident in generative AI, increasing investments, pilot programs, and deployments while expressing high satisfaction with its potential.


AI & Cybersecurity: 750 Million Attacks a Day at Amazon


Remembering Suchir Balaji: The OpenAI Whistleblower

Suchir Balaji, a principled and courageous whistleblower who exposed ethical concerns at OpenAI, argued that AI systems relying on unlicensed, copyrighted content threaten the internet’s sustainability, and his insights continue to challenge the narrative of fair use in AI.


Using AI to Understand the Copyright Controversies Arising From AI

The development of the AI-based thematic analyzer CORA demonstrates how AI, complemented by human insight, efficiently identifies key themes from thousands of public comments on generative AI copyright issues, revealing a spectrum of pro-creator and pro-AI development perspectives while highlighting the need for balanced analysis and ethical integration.


Google?—?Agents

Google’s white paper on AI agents envisions a transformative shift in business operations, highlighting agents’ ability to reason, plan, and act autonomously by integrating real-time data, dynamic tools, and advanced cognitive architectures.


Agent Governance?


Building effective agents?—?Antrophic

Effective LLM agents are built using simple, composable patterns tailored to specific use cases, with augmented LLMs as foundational building blocks, complemented by workflows like chaining, routing, parallelization, and autonomous agents when needed.




Introducing?smolagents, a simple library to build agents?—?Hugging Face

Smolagents is a lightweight library designed to build AI agents with simplicity, leveraging code-based actions and tool integration for real-world tasks, offering flexibility and compatibility across multiple LLMs, including open-source and proprietary models.


ElizaOS: Revolutionizing Autonomous AI Agents ElizaOS is an open-source protocol enabling the development of autonomous AI agents, inspired by the early ELIZA program, and is used in innovative applications like AI-led DAOs.


A very brief AI crash course in 6 bullet points?—?INSEAD Professor Theos Evgeniou

Intelligence, whether in humans or AI, lies in efficiently encoding and representing the environment, with wisdom emerging as the ability to distill vast experiences into concise, impactful insights, highlighting the shared principles between evolution and machine learning.


Can AI Tools Empower Non-Developers to Build Apps??—?INSEADer Justin Tan

Using AI tools like Replit Agent, Justin Tan built a Newsletter Summariser app, reducing newsletter processing time by hours, while highlighting the ease of natural language development, flexible integration, and rapid prototyping alongside challenges like technical foundations, tool limitations, and best use for new projects.

Full write-up?by Justin


China And AI In 2025: What Global Executives Must Know To Stay Ahead

The document highlights China’s ambitious goal to lead global AI development by 2030, emphasizing its strategic advantages in government-backed innovation, a robust AI ecosystem, real-world applications, and industry-specific approaches, while addressing challenges like technological rivalry, talent retention, and data quality.


ByteDance Unveils SA2VA: Multimodal Vision-Language Models

ByteDance’s new SA2VA models combine advanced vision-language capabilities across text, images, and videos, using multimodal encoders, refined annotation pipelines, and segmentation alignment for sophisticated AI tasks, all under an open-source license.


Google is forming a new team to build AI that can simulate the physical world

Google DeepMind is assembling a team led by Tim Brooks, previously of OpenAI, to develop AI models capable of simulating real-world environments, leveraging advancements in multimodal data and generative capabilities like the Gemini and Genie models to progress toward artificial general intelligence.


Reimagining Video Games: Ethan Mollick’s Genre-Bending Experiment

Ethan Mollick creatively reimagines classic video games as entirely new genres, from “Mario Brothers” as a 1940s industrial film to “Among Us” as reality TV, using the innovative capabilities of Veo 2.


Joint Safety Testing of OpenAI’s o1 Model

The U.S. and UK AI Safety Institutes conducted pre-deployment evaluations of OpenAI’s o1 model, revealing advancements in cybersecurity and biological tasks but noting performance gaps in software development, underscoring the importance of ongoing AI safety research.


Anthropic Research Highlights LLM Vulnerabilities

New research from Anthropic shows how easily language models can bypass guardrails through subtle, automated input manipulations, raising concerns about their robustness and security.

arxiv paper


Want to Turbocharge Your Learning?

Google’s Deep Research and NotebookLM, provide a powerful AI-driven tool for learning, offering detailed briefings, source transparency, and multi-language support to supercharge knowledge acquisition.


What China’s critical mineral ban means for the US

China’s ban on exporting critical minerals to the U.S. marks a significant escalation in the trade war, targeting materials vital to semiconductors, defense, and EVs, while leaving room for further economic retaliation in response to incoming U.S. tariffs.


The Innovation Fallacy: In the U.S.-Chinese Tech Race, Diffusion Matters More Than Invention

Jeffrey Ding argues that in the U.S.-China technological competition, the ability to diffuse and apply innovations across industries and societies is more crucial than merely inventing new technologies, shaping economic and global power dynamics.


Amazon’s Nova Models Fall Short in Chatbot Rankings

Independent rankings place Amazon’s Nova models, including the Pro version, significantly behind state-of-the-art competitors, highlighting their limited competitiveness and better value options like Gemini Flash or GPT-4o-Mini for performance-driven applications.


How Silicon Valley is disrupting democracy

Two books,?The Venture Alchemists?by Rob Lalka and?The Tech Coup?by Marietje Schaake, critique Silicon Valley’s growing power and its undermining of democracy while offering insights and regulatory roadmaps to reclaim accountability and fairness in technology’s role in society.


Europe is Not a Business Backwater

The article critiques comparisons between US and European stock markets, highlighting the misleading exclusion of AI-driven US tech companies and flawed financial metrics that obscure Europe’s strong business potential.


Public From-Scratch US and EU Companies Less than 50 Years Old with $10B+ Market Cap


Driving European Competitiveness With AI

A high-level online panel from Davos will explore how AI can bolster Europe’s global competitiveness by addressing technology gaps, enhancing skills, and fostering collaboration among businesses, policymakers, and stakeholders.


Meet the Meta AI cinematic universe

Meta is launching a star-studded AI initiative, introducing 28 AI characters, new generative AI tools, and integrations across its platforms, setting itself apart from rivals by focusing on diverse, personalized interactions rather than a singular AI assistant.


Putin orders Russian government and top bank to develop AI cooperation with China

President Vladimir Putin has directed the Russian government and Sberbank to strengthen AI cooperation with China, aiming to counter Western technology sanctions and advance Russia’s AI capabilities through partnerships with non-Western nations.


AI vs. Conspiracy Theories: A Hopeful Breakthrough?

A study from MIT and Cornell shows ChatGPT can effectively challenge conspiracy beliefs, with a third of participants shifting to accurate views after 5-minute AI conversations, thanks to AI’s patient, non-judgmental, and personalized approach.


Where can I get 7 Gigawatts of power?

Northern Alberta has granted permits for a massive, low-cost data center project powered by natural gas, requiring $70 billion in funding to develop 7 gigawatts of energy capacity.


Balancing AI Power with Sustainability

OpenAI’s o3 model showcases advanced AI capabilities but highlights significant environmental tradeoffs, consuming as much energy per task as a U.S. household uses in two months.


Big Tech Will Scour the Globe in Its Search for Cheap Energy

Big Tech is rapidly expanding data centers worldwide, like in Malaysia’s Johor, as the growing demand for generative AI drives an urgent need for cheap, reliable energy akin to a new “oil.”


Italy fines OpenAI over ChatGPT privacy rules breach

Italy fined OpenAI €15 million for breaching privacy rules with ChatGPT, citing inadequate legal basis for data use, lack of transparency, and insufficient age verification, while mandating a public awareness campaign; OpenAI plans to appeal the decision.


Blitzy: Revolutionizing Enterprise Software Development

Blitzy’s Autonomous Development Platform uses 3,000+ agents to autonomously create up to 80% of enterprise software, refine code quality, and guide the remaining 20% for human completion, offering the fastest path to deployment with a free trial available now.


AI IDE Landscape: Tools Highlighted in John Rush’s Thread

John Rush’s X thread highlights new AI-powered tools, including GitHub Spark, Google IDX, and Deco’s Webfraw.ai, showcasing advancements in agentic systems, prototyping, and full-stack app generation for developers.


Firecrawl Opens Job for AI Agents Only

Firecrawl is hiring an AI agent to autonomously create, test, and refine example applications, showcasing the platform’s potential while inspiring developers with real-world scenarios.


NHS to begin world-first trial of AI tool to identify type 2 diabetes risk

The NHS in England will trial a groundbreaking AI tool, Aire-DM, capable of predicting type 2 diabetes risk up to 13 years early using ECG data, potentially enabling targeted preventive care and reducing future health complications.


Build a Large Language Model (From Scratch)

Build a Large Language Model (From Scratch)?by Sebastian Raschka offers a step-by-step guide to creating, training, and fine-tuning your own LLM, providing a practical, hands-on approach for Python users to understand and customize generative AI systems.


How Hallucinatory A.I. Helps Science Dream Up Big Breakthroughs

AI-generated hallucinations, often criticized for creating false information, are proving valuable in science by inspiring innovative ideas and breakthroughs in areas like cancer research, drug design, and weather phenomena.


Co-STORM- Get a Wikipedia-like report on your topic with AI?—?Stanford

Co-STORM is an AI research prototype that facilitates the creation of Wikipedia-like reports by supporting collaborative, multi-perspective discussions between participants, experts, and AI, enabling interactive knowledge generation and report synthesis.


Elon Musk gets new support in growing lawsuit against OpenAI

Elon Musk is intensifying a legal battle against OpenAI, seeking to block its transition to a for-profit model and accusing it of unfairly limiting investors from backing rivals like his own AI startup, xAI, while gaining support from prominent tech investors, researchers, and advocacy groups.


INSEADer Lutz Finger Prediction #1: No AGI in 2025

Lutz Finger predicts that AGI (Artificial General Intelligence) will not emerge in 2025, as current AI systems like OpenAI’s o1 rely on statistical patterns and lack the adaptability, reasoning, and real-world understanding needed for true general intelligence.


Lutz Finger Prediction #2: 2025 Will Be the Year of Search

Lutz Finger predicts that 2025 will be the year search technologies dominate, as Large Language Models (LLMs) revolutionize how we interact with information, creating fierce competition among tech giants and startups?—?but Google’s deep expertise and data advantage ensure it remains a formidable player.


Lutz Finger Prediction #3: AI Agents Will Define 2025

Lutz Finger predicts 2025 will be the year of AI agents revolutionizing enterprise workflows, as businesses move beyond basic GPT wrappers to develop integrated, reliable, and dynamic agentic systems that simplify and enhance processes across industries.?


Lutz Finger Prediction #4: Multimodal AI Will Transform Enterprises in 2025

Lutz Finger predicts that multimodal AI will revolutionize industries in 2025 by integrating diverse data types?—?text, images, audio, and more?—?enabling powerful applications in healthcare, eCommerce, and beyond, providing richer insights and transforming enterprise workflows. (Stay tuned for more predictions!)


Lutz Finger Prediction #5: AGI Hype and the 2025 Data Battle

Lutz Finger critiques Sam Altman’s claim that AGI is near, emphasizing that current AI, including OpenAI’s o1, relies on iterative reasoning rather than true understanding, while predicting that 2025 will intensify competition among tech giants for data dominance and raise the need for workforce education in AI.


Lutz Finger Prediction #6: AI Agents Enter the Workforce in 2025

Lutz Finger predicts 2025 will see AI agents integrated into workplaces, transforming tasks with specialized workflows, while tech giants like Microsoft, Google, and Salesforce vie to dominate the market through superior user access and data integration, leaving OpenAI challenged to compete on scale and pricing.


2025: A Breakthrough Year for AI Prototyping and Learning?—?Andrew Ng

AI is revolutionizing software development by making it faster and easier to build prototypes, enabling creative applications like flashcard generators or financial monitoring tools, while encouraging everyone to embrace coding, continuous learning, and the joy of building.


Audrey Tang: AI That Unites Us

Audrey Tang envisions AI as a force for empathy and collaboration, urging a shift from engagement-driven algorithms to prosocial, inclusive systems that foster social cohesion, democratic values, and collective problem-solving.


25 Predictions on Where AI Will Be at the End of 2025?—?Gary Marcus?

Despite ongoing advancements and hype, AI in 2025 will remain limited in achieving artificial general intelligence, solving complex tasks, and demonstrating reliability, with modest profits, regulatory gaps, and overhyped technologies like humanoid robots and AI agents, while challenges such as hallucinations, reasoning errors, and power consumption persist, and adoption of cutting-edge systems like GPT-5-level models remains uncertain.


7 Predictions for 2025?—?The Year AI Fights Back Against Scams?—?Sardine

2025 will see AI revolutionize fraud prevention, transforming risk operations with intelligent agents, driving cross-industry collaboration, addressing regulatory demands for model validation, and pushing towards integrated fraud and compliance systems, as financial institutions combat increasingly sophisticated AI-driven scams.


Predictions + Scorecard, 2025 January 01?—?Rodney Brooks

Rodney Brooks reflects on the reality of overhyped predictions in robotics, AI, self-driving cars, and space exploration, offering insights from his predictions and emphasizing the gap between technological promises and practical deployment.


The 2025 AI Engineer Reading List

This curated list of 50 papers, blogs, and models across 10 AI engineering fields?—?including LLMs, multimodal AI, RAG, and fine-tuning?—?offers essential resources for AI engineers to stay updated on cutting-edge advancements and practical applications in 2025.


OK, I can partly explain the LLM chess weirdness now

Despite being smaller and older, OpenAI’s GPT-3.5-turbo-instruct uniquely excels at chess compared to newer models, likely due to high-quality chess training data and differences in instruction tuning and chat interfaces, highlighting the complexity of optimizing LLMs for specialized tasks.


The Rise of AI Dev Tools: Transforming App Development

Colin Matthews outlines three categories of AI dev tools?—?chatbots, cloud dev environments, and local dev assistants?—?and provides practical guidance on choosing the right tool, solving prototyping challenges, and maximizing productivity for building apps in minutes.


Apple Intelligence Now Takes Up Nearly Twice as Much Storage as It Did at Launch

Apple Intelligence, despite low user enthusiasm, now takes up 7GB of device storage?—?nearly double its launch size?—?due to on-device AI processing for new features like image generation and custom emojis, raising concerns about storage, battery drain, and overall value.


Zoom will let AI avatars talk to your team for you

Zoom will soon allow users to create AI avatars that can record and deliver personalized video messages, part of its AI Companion add-on, which includes features like integration with productivity apps and custom meeting summaries.


The Global AI Research Agenda & AI in Global Development Playbook

U.S. Secretary of State Antony Blinken announced the official release of the?Global AI Research Agenda?and?AI in Global Development Playbook?—?two comprehensive and complementary approaches that guide future research on Artificial Intelligence (AI) and its use in advancing UN Sustainable Development Goals.

NotebookLM summary


AIR-BENCH 2024: A Safety Benchmark Based on Risk Categories from Regulations and Policies

AIR-BENCH 2024 introduces a comprehensive AI safety benchmark grounded in regulatory frameworks and policies, encompassing 5,694 prompts aligned with 314 granular risk categories, to evaluate language models’ safety performance and address gaps in existing benchmarks by providing insights into real-world regulatory alignment and safety measures.


Should you trust an AI-assisted doctor? I visited one to see.?—?Paywall

A columnist explores the benefits and challenges of AI-assisted doctors, raising questions about their accuracy, bias, and reliability in healthcare.


AI on a Calculator: Achieving the Impossible

In a 56-hour train journey, the author ported a neural network to a TI-84 Plus CE graphing calculator, achieving over 95% accuracy in handwritten digit recognition despite hardware limitations, showcasing ingenuity in machine learning on resource-constrained platforms.


SimpleQA: A Benchmark for LLM Factual Accuracy

OpenAI’s SimpleQA tests LLMs on 4,326 short, fact-seeking questions across diverse topics, revealing challenges in accuracy, overconfidence in wrong answers, and opportunities for improvement, with results soon available on Hugging Face.


A Framework for Consciousness

Francis Crick and Christof Koch propose a framework to explain visual consciousness by identifying neural correlates of consciousness (NCC) through competing neural assemblies, combining experimental approaches and theoretical insights to address the fundamental problem of qualia.


Winning Tech Means Winning Everything

Dominating technology is essential for success across industries, as it encompasses not just innovation in AI and transportation but also critical sectors like finance, media, and defense, a strategy that China clearly grasps.


Are Language Models Actually Useful for Time Series Forecasting?

Ablation studies reveal that large language models (LLMs) offer no clear advantage in time series forecasting, as simpler attention-based methods often perform as well or better, questioning the utility of LLMs in this domain despite their computational cost.

Commentary


AI and Auditing: Transforming Practices and Evolving Standards?—?(French)?—?INSEADer Stéphane Bellanger

This article explores how artificial intelligence is revolutionizing auditing by enhancing decision-making, fraud detection, and service personalization, while addressing challenges like transparency and integrating AI into ethical and professional standards.


Sparse autoencoder

Andrew Ng’s lecture notes introduce sparse autoencoders as a method to automatically learn efficient feature representations from unlabeled data, addressing the limitations of manually engineered features in supervised learning and leveraging neural networks with backpropagation for scalable, domain-agnostic applications.


How AI revolution could help benefits appeals and landlord disputes

AI is transforming the legal field by enhancing access to justice through tools that streamline case preparation, analyze evidence, and reduce costs, while emphasizing the need for human oversight to ensure fairness and accuracy.


AI and the Future of Digital Public Squares

AI and LLMs hold transformative potential to enhance digital public squares by facilitating collective dialogue, bridging divides, moderating content, and ensuring authenticity, while also addressing challenges like polarization, inclusivity, and trust.


Why the World Needs Lazier Robots

Embracing “lazy robotics,” which prioritizes energy-efficient actions and selective data processing, can significantly reduce the environmental impact of robots and AI while maintaining functionality and addressing the growing energy demands of automation.


In a First, Surgical Robots Learned Tasks by Watching Videos

Surgical robots have achieved human-level proficiency by learning tasks through video training, including the ability to correct their own mistakes during procedures, marking a breakthrough in robotic-assisted surgery.


What is the best LLM for RAG and grounded long-form inputs?

Google DeepMind’s FACTS benchmark evaluates LLMs for factually accurate, grounded long-form responses, with Gemini 2.0 Flash outperforming others at 83.6% accuracy, emphasizing grounding over creativity, and covering diverse domains like medical, legal, and tech.


Unit X: Bridging Silicon Valley and Pentagon Innovation (Book)

Unit X unveils the groundbreaking partnership between Silicon Valley and the Pentagon, showcasing how cutting-edge technology is transforming modern warfare, as narrated by the two key leaders who spearheaded this pivotal shift.

CNBC Segment on Unit X


AI Agent Revolutionizing Job Applications

“Browser Use” is an AI tool that automates job applications by analyzing your CV, finding matching jobs, and applying for them autonomously, showcasing the transformative potential of AI agents in reshaping the recruitment and hiring market.


Microsoft Phi-4: A New Benchmark in STEM-Focused LLMs

Microsoft Phi-4, a 14B parameter LLM released on Hugging Face under an MIT license, excels in reasoning, math, and code generation, outperforming GPT-4o and rivaling 70B models with its STEM-focused QA capabilities, high-quality training data, and advanced safety alignment.


AI, Investment Decisions, and Inequality?—?MIT Sloan

Generative AI enhances investment decision-making overall but risks widening performance gaps, as personalized summaries benefit sophisticated investors more, highlighting a tradeoff between accessibility and precision.


Breaking Ground at NeurIPS 2024: Advancing Physics-Informed AI and Beyond

Excited to present our cutting-edge research at NeurIPS 2024, showcasing breakthroughs in physics-informed learning for stochastic processes, conditional operator estimation, Gibbs algorithm generalization, operator learning for reinforcement learning, and novel contributions to RNN theory and ML potentials at key workshops.


Yuma?—?Get the most advanced AI Agents for unmatched Customer Service

Yuma delivers leading automation rates for E-commerce while improving? support accuracy, consistency, and cost-efficiency.


Learning Theory from First Principles?by Francis Bach Francis Bach’s new book offers a fresh, rigorous approach to understanding learning theory through first principles, simplifying complex mathematical concepts while bridging theoretical results with practical applications in machine learning, making it an invaluable resource for students and researchers alike.


Kant, Popper, and Vapnik: Foundations and Limits of AI

Reflecting on Vapnik’s foundational connections between Kant, Popper, and learning theory, this discussion revisits the philosophical and scientific underpinnings of machine learning, exploring its relevance to modern AI challenges like hallucinations, its limitations, and what is still missing in advancing AI’s true potential.


Robots That Learn: UC Berkeley Course Materials?—?Jitendra MALIK

UC Berkeley’s “Robots That Learn” course, featuring lecture videos and comprehensive materials, is now available online, offering insights into advanced robotics and learning systems with expert guidance.


In-Context Learning of Representations in LLMs

New research reveals that large language models (LLMs) dynamically form “in-context representations” that align with task structures, reflecting the real-world data generating processes, offering insights into their internal reasoning capabilities.


Quantum Economy Landscape in Saudi Arabia

Saudi Arabia is advancing its quantum technology ambitions through strategic investments, partnerships, and education initiatives, aiming to become a global leader in quantum innovation while addressing challenges in talent development and infrastructure to align with Vision 2030 goals for economic diversification and technological advancement.


Alien Recombination: Exploring Concept Blends Beyond Human Cognitive Availability in Visual Art

Researchers propose the Alien Recombination method, using AI to transcend human cognitive limitations by generating novel concept combinations in visual art, revealing unexplored creative possibilities beyond cultural and contextual constraints.


Announcing 150M developers and a new free tier for GitHub Copilot in VS Code

GitHub now offers a free tier of GitHub Copilot in VS Code, allowing 150M developers to access up to 2,000 code completions and 50 chat messages per month, with support for models like Anthropic’s Claude 3.5 Sonnet and OpenAI’s GPT-4o.


The Circle: A Cautionary Tale on AI, Privacy, and Tech Ethics?—?Film Recommendation

This 2017 film explores the dangers of unchecked AI and tech monopolies, raising critical questions about transparency, surveillance, and the ethical responsibilities of companies


Ambitious Tech for 2025: Mind-Reading Goals with Omi

Nik Shevchenko introduces Omi’s vision to “read the mind,” starting with device activation via a brain interface module launching in Q2 2025, a simpler step toward achieving fully brain-powered interactions.


Book on AI, Behavioral Science, and Digital Persuasion

Sandra Matz’s book explores the world of AI-driven psychological profiling, algorithmic nudging, and digital persuasion, offering essential insights into the powerful and sometimes unsettling intersection of data science and human behavior.


The Crowd: A Study of the Popular Mind?—?1895

Gustave Le Bon’s 1895 classic,?The Crowd: A Study of the Popular Mind, explores the psychology of crowds, asserting that individuals in groups act impulsively, lose rationality, and are swayed by emotions and leaders, influencing civilization through their destructive or transformative power.


UK Delays Gene Editing Plans in Push for Brexit Reset

The UK’s gene-editing plans face delays as Brussels warns they conflict with post-Brexit border agreements, sparking concerns over Europe’s regulatory competitiveness in critical tech sectors.


AI Videos of the Last Few?Weeks

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

AI Semiconductor Landscape feat. Dylan Patel

  1. NVIDIA’s Competitive Edge: NVIDIA dominates AI workloads through its integration of advanced hardware, software, and networking, supported by a rapid annual innovation cycle.
  2. Synthetic Data and Scaling: Synthetic data generation and inference-time compute are emerging as critical factors for scaling AI models as pre-training faces diminishing returns.
  3. Rising Competition: Companies like AMD, Amazon, and Broadcom are challenging NVIDIA with cost-effective chips and systems, though they lack NVIDIA’s holistic system-level capabilities.
  4. Hyperscaler Investment Surge: Major tech firms like Microsoft, Google, and Meta are massively increasing CapEx for AI infrastructure, building larger data centers and clusters to meet scaling demands.
  5. Memory and Cost Dynamics: Advanced reasoning models are driving unprecedented memory demands, increasing costs in high-bandwidth memory (HBM) and creating challenges for cost-efficient AI deployments.


The Revolutionary Potential of AI with CSAIL Professor Manolis Kellis

  1. AI’s Progress Towards AGI: Professor Manolis Kellis argues that AI has already surpassed many definitions of Artificial General Intelligence (AGI), excelling in intellectual tasks and moving towards physical capabilities, redefining how intelligence is perceived.
  2. Transformational AI Tools: Kellis envisions a future where multimodal AI systems integrate language, visual, and structured data, offering businesses and researchers tools to navigate and understand vast landscapes of information with transparency and attribution.
  3. Impact on Medicine and Science: AI’s potential to revolutionize medicine is highlighted, with applications in personalized therapies, drug development, and understanding complex biological systems, leveraging AI’s ability to reason about data beyond human limitations.
  4. Ethics and Responsibility: Kellis emphasizes that AI is a tool, neither inherently good nor bad, and that users must bear responsibility for its applications, drawing parallels to other tools like hammers or vehicles.
  5. Future Innovations and Accessibility: Introducing projects like MantisAI, Kellis highlights the importance of creating human-centric interfaces for data science, allowing users to interact with and manipulate data intuitively, democratizing AI’s capabilities.


Microsoft CEO’s Shocking Prediction: “Agents Will Replace ALL Software”

  1. Agent-Driven Future: Microsoft’s Satya Nadella predicts the decline of traditional business applications in favor of AI agents directly interacting with databases, marking a significant shift in software development.
  2. Simplification of Business Logic: Business logic, traditionally handled by applications, will be compressed into agents capable of executing tasks by directly engaging with core data, eliminating the need for complex interfaces.
  3. Impact on SaaS Industry: This transformation could render traditional SaaS models obsolete, reducing demand for developers and potentially reshaping the software industry landscape.
  4. AI-Native Tools: Tools like Excel and other business applications are evolving into AI-enhanced platforms, where agents handle data analysis and decision-making tasks without requiring traditional software interfaces.
  5. Strategic Realignment: Companies must adapt to this new paradigm by adopting agent-driven architectures or risk obsolescence, as seen with Microsoft’s investment in AI-powered tools like Co-Pilot.


Why Amazon, Microsoft, Google And Meta Are Investing In Nuclear Power

  1. Massive Energy Needs: Tech giants like Amazon, Microsoft, Google, and Meta are investing in nuclear power to meet the escalating energy demands of AI-driven data centers, which require power comparable to entire cities.
  2. Sustainability and 24/7 Power: Small modular reactors (SMRs) are gaining traction due to their ability to provide reliable, emissions-free energy, addressing the intermittency issues of solar and wind power.
  3. Strategic Investments: Companies are making significant investments, such as Amazon’s $500M nuclear projects and Microsoft’s revival of Three Mile Island, to ensure energy independence and future scalability.
  4. Overcoming Opposition: Despite public skepticism fueled by past nuclear accidents, tech companies’ financial resources and long-term vision are helping to reshape perceptions and address cost challenges.
  5. Driving Energy Innovation: By championing nuclear power, these companies are positioning themselves as leaders in the global energy transition, aligning technological growth with sustainable energy solutions.


The Heist: A Text-to-Video Showcase?(AI-creation)

The Heist?demonstrates the power of Google Veo 2’s text-to-video technology, with all sound design, editing, and music crafted by the creator, teasing exciting projects ahead for next year.


Palmer Luckey, Founder of Anduril Defense Industry Disruptor?—?President Speaker Series

  1. Disrupting the Defense Industry: Palmer Luckey founded Anduril Industries in 2017 to modernize U.S. defense capabilities by combining artificial intelligence with cutting-edge hardware. His company focuses on rapid product development and deployment, with applications in U.S. military branches and active conflicts like the Russia-Ukraine War.
  2. Non-Traditional Background: Luckey’s unconventional journey began with being homeschooled and attending community college at 14. He left university to develop the Oculus Rift, revolutionizing virtual reality, before transitioning to the defense sector.
  3. Ethical Defense Innovation: At Anduril, Luckey emphasizes a mission-driven approach, blending private-sector agility with national security needs. He critiques traditional defense contractors for inefficiencies and positions Anduril as a proactive product company focused on impactful defense technologies.
  4. Philosophy on Freedom and Technology: Luckey advocates for leveraging advanced AI and autonomous systems to ensure national security while preventing evenly matched conflicts. He emphasizes “peace through strength” and the ethical responsibility of using technology to protect democratic values.
  5. Broader Technological and Business Vision: Beyond defense, Luckey explores ideas like synthetic foods and nuclear energy. He also launched a luxury Game Boy Color clone, blending nostalgia with modern engineering, and supports innovation across sectors to tackle pressing global challenges.


Fun


Past editions?—?AI NEWS YOU?MISSED

View list


SUBSCRIBE TO GET UPDATES IN YOUR?MAILBOX

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

Constance d'Aspremont

CEO Ethics & Boards | ??? Governance | ?? Sustainability | ?? Strategy | ?? Data Analytics | Board Member | Angel Investor

1 个月

Thanks Robert Maciejko!

Justin Tan

AI Transformation | AI Educator | Speaker

1 个月

Great wrap up as always, Robert!

Thorsten L.

Driving business transformation with AI agents and workflow automation. At InnovareAI, we help companies automate tasks, reduce costs, and achieve measurable growth.

1 个月

Robert Maciejko, the rapid evolution of ai technology presents remarkable opportunities for strategic growth across multiple sectors. would you agree?

要查看或添加评论,请登录

Robert Maciejko的更多文章

  • AI NEWS YOU MISSED??#57

    AI NEWS YOU MISSED??#57

    Top 5 stories of the week 1?? OpenAI launches GPT-4.5 2?? Claude 3.

    1 条评论
  • AI NEWS YOU MISSED ?#56

    AI NEWS YOU MISSED ?#56

    Top 5 stories of the week 1?? Elon Musk’s Grok 3 released 2?? ChatGPT 5 in May? 3?? Deep Research x 3 + 1 4?? Microsoft…

    10 条评论
  • AI NEWS YOU MISSED ?#55

    AI NEWS YOU MISSED ?#55

    Top 5 stories of the week 1?? No AI risk or job impact?—?Vance 2?? OpenAI AI roadmap 3?? Gemini leads. Nobody notices.

    4 条评论
  • DeepSeek: Myths, Realities & Controversies

    DeepSeek: Myths, Realities & Controversies

    DeepSeek, a Chinese AI startup, dominated global news this week with its AI models, reportedly developed using Nvidia…

    11 条评论
  • AI NEWS YOU MISSED ?#54

    AI NEWS YOU MISSED ?#54

    Top 5 stories since January 9 1?? DeepSeek new OpenSource AI stuns 2?? OpenAI launch agent Operator 3?? Trump announces…

    2 条评论
  • AI NEWS YOU MISSED??#52

    AI NEWS YOU MISSED??#52

    Where we are on AI, where we are going In the past weeks, Google, OpenAI, and other developers have hit us with…

  • AI NEWS YOU MISSED ?#51

    AI NEWS YOU MISSED ?#51

    Top 5 stories of the week 1?? OpenAI’s “12 Days of OpenAI” midpoint 2?? Gemini 2.0: Advancing Agentic AI 3?? Willow:…

  • AI NEWS YOU MISSED ?#50

    AI NEWS YOU MISSED ?#50

    Top 5 stories of the week 1?? 12 Days of OpenAI Starts With o1 Model 2?? Amazon Nova Models Unveiled 3?? Anduril &…

    3 条评论
  • AI NEWS YOU MISSED ?#49

    AI NEWS YOU MISSED ?#49

    It’s Thanksgiving week in the US! After a recent surge of AI news, this week has been quieter on the headlines…

  • AI NEWS YOU MISSED ?#48

    AI NEWS YOU MISSED ?#48

    Top 5 stories of the week 1?? Microsoft unleashes agents & more 2?? NVIDIA’s 94% growth?—?mixed reactions 3?? Google…

    2 条评论

社区洞察

其他会员也浏览了