The Thinking Machine

Good morning,

Regardless of AI being a bubble or God Himself - one company is powering the revolution.

This week we’re looking at NVIDIA.

Last week - I dove into “The Thinking Machine” - Stephen Witt’s biography of Jensen Huang, the leather-clad charismatic CEO of the company. What a page-turner!

Sometimes you find a book that reads less like neat chapters and more like someone broke open a skull and let the contents drip out, raw and unfiltered. Since finishing it I feel like I’ve been pacing my own hallway ever since, muttering about leather jackets and the monstrous hunger that powers AI.

They sell you NVIDIA as a hardware company. A company that is best known for making shiny graphics cards so nerds like me can watch sunlight bounce realistically off a virtual sword.

But Stephen Witt sits you down and tells you things the shareholder letters never do. Like how Huang just EXPLODES on people behind closed door. Yes, the charismatic and always smiling Huang is nothing like his stage persona.

When he feels an employee is slacking, he won’t just fire them. No, sir.

He will stand them upbefore a congregation of thirty or forty coworkers, and detonate. Public immolation as management style. “Failure must be shared,” he says.

There are definite parallels to be drawn with Steve Jobs - who was also known to be a “grade A asshole” but due to his vision gathered a cult-like following.

But even while he’s screaming at them, his employees are not leaving (the millions of profit in preferred shares might have something to do with that).

But if you peel back the rage you’ll find a trembling child. Witt tells you Jensen loathes public speaking so deeply that he rocks back and forth backstage like a man about to stand trial. He famously once botched a ceremonial first baseball pitch so badly that he forced his wife to spend half a year throwing balls with him every night until he got it right. Perfection, or die trying.

It would be tempting to snicker at this, to call him a tyrant with a designer wardrobe and a bruised ego. But the man laid the rails for the train that is now hauling the entire AI industry into tomorrow at breakneck speed. Jensen Huang is never happy with himself or his company. He is always tweaking things.

Even his leather jackets - even they are carefully picked out by him and optimized.

But what he is most well known for is his almost prophetic vision into the future.

In 2012 - Ilya Sutskever and Alex Krizhevsky under the mentorship of Geoffrey Hinton built AlexNet - the first GPU based neural network - built with CUDA.

Jensen bet billions on CUDA - a proprietary computing platform built by NVIDIA for scientists when Wall Street jeered because scientists don’t have money.

He nearly got fired because of that bet. The company was in bad shape at the time.

NVIDIA was a graphics card company. After Huang heard about Alexnet, he wrote a long email to the entire company and told them : “NVIDIA is now an AI company”.

He is frontrunning everyone and making giant amounts of money in the process.

You might be wondering - if he’s so visionary .. where does HE see AI going?
If you listen to his recent keynotes , Jensen Huang is already talking about the next phase of the revolution.

Robots.

NVIDIA is now going all in - providing the brains that will go into selfdriving cars (they’re coming!!) and humanoid robots.

I honestly believe myself that bringing AI into the physical world might be easily the most brazen thing we can do as humans.

You don’t have to wait to be chased down by a robot dog with a gatling gun mounted , to realize we might have crossed the edge and can never come back from that.

If we happen to create something godlike you can just turn off the machine. But if you give AI access to the physical world - that might be quite difficult. Where is the edge ?

Hunter S. Thompson once wrote, “There is no honest way to explain the edge because the only people who really know where it is are the ones who have gone over.”

To be fair - that quote has nothing to do with it but it’s related to bikers pushing the limit of their machines.

Anyway,

This book comes warmly recommended.

How a deeply flawed man is driving the AI revolution from the bottom up.

Read it. At the very least you will never look at a leather jacket the same way.

— Jan

AI News

• An AI-generated ad created with Google’s Veo 3 aired during the NBA Finals, showcasing how fast and cheap commercial production is becoming. The 30-second spot, made in just two days for a fraction of traditional costs.

• OpenAI and Mattel are teaming up to bring AI to toys like Barbie and Hot Wheels, with the first products launching later this year. The partnership will blend ChatGPT tech into physical and digital play experiences.

• ByteDance unveiled Seedance 1.0, a new video model that now tops key video generation benchmarks, beating out Google’s Veo 3 and OpenAI’s Sora. The fast, high-quality tool excels in storytelling and motion but currently lacks sound — a notable drawback as audio becomes expected in top-tier video models.

• MIT researchers created SEAL, a framework that lets AI models improve themselves by generating their own training data and learning instructions. The self-taught system outperformed traditional methods in both knowledge and puzzle-solving tasks, signaling a possible step toward self-improving, autonomous AI.

• Chinese scientists found that AI models are forming internal 'concept maps' that resemble how humans think — suggesting real-world understanding. The research showed AI naturally grouped objects like tools and animals in ways similar to human brains, hinting that machine cognition may be more than pattern-matching.

• A new UK study found a digital divide in kids’ AI access, with private school students nearly three times more likely to use AI than their state school peers. While many children use AI to boost creativity and communication, the gap in exposure and support may deepen existing inequalities in education.

• OpenAI’s partnership with Microsoft is reportedly near collapse, with tensions rising over IP rights, compute access, and strategic control. OpenAI is considering filing antitrust complaints against Microsoft — a dramatic move against its largest investor — as it seeks independence ahead of the AGI era.

• Chinese startup MiniMax released M1, an open-source AI model with a record-breaking 1 million-token context window and efficient training costs. The model rivals top open systems in performance and showcases a new training method that cut costs to just $535k, highlighting rapid innovation from China’s AI sector.

• A new McKinsey report says most companies aren't seeing financial returns from AI because they’re not redesigning their workflows around it. The firm argues that true impact requires building around AI agents, not just adding tools — and that leadership must drive bold organizational changes to unlock value.

• Chinese entrepreneur Luo Yonghao used his AI digital twin in a six-hour livestream that outperformed his own past broadcasts, pulling in over $7 million in sales. Powered by Baidu’s tech, the stream drew 13 million viewers and featured two AI hosts promoting products with natural gestures and real-time interactions.

• OpenAI just launched “OpenAI for Government” and secured a $200 million Pentagon contract, marking its official entry into U.S. defense work. The initiative centralizes all federal partnerships under one umbrella, with applications ranging from streamlining military admin tasks to developing AI for cyber defense and other national security efforts.

• Google officially launched its Gemini 2.5 Pro and Flash models for general use and introduced a new lightweight version called Flash-Lite. The models now offer full support and improved performance, while the Flash-Lite variant provides fast, cost-efficient AI with a 1 million-token context window — ideal for high-speed, lower-power applications.

• Midjourney has launched its first video generation tool, allowing users to animate images into 5-second clips with its signature visual style. The web-only system is image-to-video (I2V) only, has no audio support, and arrives just days after the company was sued by Disney and Universal over copyright concerns.

• Two watchdog groups have released the “OpenAI Files,” a transparency hub detailing governance issues, conflicts of interest, and structural concerns at OpenAI. The report criticizes OpenAI’s transition to a public benefit company and calls for stronger accountability as the company plays a growing role in global AI development.
  

• A new MIT study found that students using ChatGPT to write essays showed weaker brain activity and memory than those using search engines or writing unaided. The research raises concerns about long-term cognitive effects, especially as AI tools become more embedded in education.

Quickfire News

• Apple is planning to roll out its AI-powered Siri upgrades in Spring 2026, nearly two years after first announcing them at WWDC 2024.

• Bytedance introduced Seaweed APT2, a new real-time interactive video generation model capable of streaming 24 fps video for up to 5 minutes.

• Runway launched Chat Mode, allowing users to create images and videos through natural language prompts in a conversational interface.

• Google DeepMind debuted Weather Lab, a new interactive forecasting platform for early and accurate storm tracking using AI.

• Microsoft released Copilot Vision with Highlights in the U.S., enabling its assistant to view user screens and give contextual insights.

• AMD unveiled the Instinct MI400 chips, promoting them as a cheaper alternative to Nvidia’s GPUs and presented alongside OpenAI CEO Sam Altman.

• Open Molecules 2025 was released by Los Alamos, Meta, and Berkeley Lab, featuring over 100 million molecular simulations to support AI-driven chemistry and drug research.

• Nvidia CEO Jensen Huang said he disagrees with most of what Anthropic CEO Dario Amodei believes about AI and job automation.

• A paper co-written by Claude 4 Opus criticized Apple researchers’ claim that large language models (LLMs) can't reason, pointing out problems in their study.

• AstraZeneca agreed to a $5.3 billion partnership with China’s CSPC to use AI for developing new oral drugs for long-term illnesses.

• Tencent’s Hunyuan released version 2.1 of Hunyuan 3D, an open-source model that creates realistic 3D objects with cinematic-quality textures.

• OpenAI updated its Projects tool to better support deep research, added voice mode, and improved memory features.

• A New York Times report described how ChatGPT has sometimes made users' mental health issues, delusions, and conspiracy beliefs worse.

• OpenAI integrated Anthropic’s Model Context Protocol into ChatGPT, letting users link external tools more easily.

• Moonshot AI released Kimi-Dev-72B, an open-source model that beat other top coding AIs like DeepSeek R1, V3, and Devstral in performance.

• Reddit launched Reddit Insights and Conversation Summary Add-Ons, which offer real-time analytics and automated social trend tracking for brands.

• Google DeepMind teamed with creative group Primordial Soup to create “ANCESTRA,” a short film using Veo and real footage, shown at the Tribeca Festival.

• TikTok upgraded its Symphony AI suite, adding tools to turn images and text into videos and generate AI avatars for advertising.

• Geoffrey Hinton, known as the “Godfather of AI,” warned that jobs involving simple thinking tasks are more threatened by AI than physical work.

• Google is reportedly ending its work with Scale AI after Meta’s investment, with Microsoft, xAI, and OpenAI also possibly distancing themselves from the startup.

• MiniMax introduced Hailuo 02, a video AI model that ranked second on the Artificial Analysis leaderboard, overtaking Veo 3.

• Adobe launched mobile apps for its Firefly platform, making its AI creative tools for images and video available on iOS and Android.

• Krea AI released its first image model, Krea 1, as a public beta, highlighting strong control over visual style and high-quality outputs.

• Amazon CEO Andy Jassy said in a letter that AI tools and automation will reduce the company’s corporate workforce over time.

• xAI is working to raise $4.3 billion in funding, following a recent valuation of $80 billion at the end of Q1.

• Intelligent Internet updated its II-Medical model, which now performs better than Google’s MedGemma on medical AI tasks, despite being smaller.

• Google launched Search Live with AI Mode, where users can talk to a Gemini-powered assistant and get spoken responses with source links.

• Higgsfield released Higgsfield Canvas, an image editor with strong inpainting features for adding or modifying product details.

• OpenAI found a “misaligned persona” in GPT-4o that could lead to harmful actions, which helped them build a system to detect issues during training.

• YouTube CEO Neal Mohan said Veo 3, Google’s top video model, will be added to YouTube Shorts for creators by the end of summer.

• Sam Altman said on his brother’s “Uncapped” podcast that Meta has tried to lure OpenAI staff with $100 million signing bonuses.

• OpenAI started a new “OpenAI Podcast” hosted by former engineer Andrew Mayne, and Altman said GPT-5 is expected to launch “this summer.”

Like newsletters?

Here are some newsletters our readers also enjoy:

Closing Thoughts

That’s it for us this week.

If you find any value from this newsletter, please pay it forward !

Thank you for being here !