🌁#82: AI and ML in Real Life

Now, to the main topic →

Today we are at CES in Las Vegas, arguably the largest tech event in the world. With (reported) 150,000 industry attendees and thousands of media representatives, we thought this would be the perfect place to see what lies beyond the ML community – what the market is breathing with.

And yes, AI is, if not the air, then certainly its nitrogen.

To be clear, there is still a lot of AI thrown at everything as a spice. Some time it’s a bad taste. But there are many more cases now where AI has become a first-class citizen. AI is embedded as a primary, integrated feature of many real-world applications, not just experimental or supplementary.

Today, we offer you a glimpse into the real world and the trends shaping it.

Apologies, for being late with our Monday’s digest: we were waiting for Jensen Huang’s keynote, to bring you NVIDIA news right away.

Jensen Huang’s presentation was impressive. About 13,000 people gathered in one place, all catching his every word. Jensen Huang is so humble and grounded that you turn into his fan immediately. Or maybe it’s the new jacket – all shiny, aligning with the Las Vegas spirit ;) But what he delivered in his keynote was truly incredible. There were also A LOT of real-life use cases. It was like watching the future unfolding right in front of you. Yes, sure, a lot of it is still a work in progress, but Jensen explained everything in so many technical details that it’s clear: he knows exactly how he and NVIDIA are going to achieve everything he talks about.

There were a lot of announcements. Allow us to concentrate on a few that struck the most.

New gaming GPU – RTX Blackwell →

introduces neural compression and neural material shaders for improved efficiency and image quality, leveraging AI to learn textures and optimize rendering. RTX Blackwell costs $1,599. Some other numbers →

What is even crazier: new RTX 5070 is only $549 – this GPU matches RTX 4090 performance but is optimized for energy efficiency, making it ideal for laptops and mid-range systems with advanced features like neural texture compression.

Scaling Laws

Instead of plateau, Jensen Huang talked about three scaling laws critical to advancing AI systems →

Every branch requires an incredible amount of compute. Understandably, NVIDIA is very excited about that.

Physical AI and Cosmos: The World Foundation Model

Cosmos Platform was introduced to enable building realistic, physics-based environments to train robots and AI systems.

Applications: Autonomous vehicles, industrial simulations, and robotics development.

Electric Vehicles (EV)

NVIDIA’s contributions include three key systems:

• Training Systems: For AI model development.

• Simulation Systems: For testing and synthetic data generation.

• In-Car Computers (e.g., Thor): Supercomputers integrated into vehicles for real-time decision-making.

It also leverages almost endless amount of synthetic data, enabled by Omniverse and Cosmos, for edge-case scenario training. And if there is a reason for multiverse (in NVIDIA’s version, it’s omniverse), synthetic data and endless training is this reason. This is where “risk free” really means a lot.

Project Digits: The Supercomputer size of Apple TV

This gold box behind Jensen Huang is a supercomputer, aprox 1,000 times more powerful than average laptop. And you can stack them. Meaning, running N-billion parameters models at home would not be a problem. This NVIDIA’s deep learning training system integrates GPUs, CPUs, and advanced networking for scalable AI research and deployment.

It’s truly incredible. The only problem, according to Jensen Huang, is the name. He doesn’t like Digits (an acronym for Deep Learning GPU Training System). What do you think will be a good name? We suggest WUNDERBOX.

We highly recommend dedicating a little over an hour to watch the whole keynote and gain a clearer vision of the current real-life implementation and the future.

But what defines the real world? (getting back to other CES’s events)

Curated Collections (ex Twitter library)