NVIDIA flexes muscles at CES 2025

Along with: Sam Altman's Manifesto to Start the Year

Hey there, 

Nvidia has made some big updates during the CES 2025, but let’s be real- Project DIGITS stole the show. 

Imagine a desktop the size of a Mac Mini, packing the power to run AI models with up to 200 billion parameters. That’s a game-changer.

This might just spark the rise of next-gen AI desktops for data scientists. Who knows, we could soon see a wave of specialized, compact AI models everywhere. Exciting stuff, right?

Now, let’s check out some of the updates this week.

What would be the format? Every week, we will break the newsletter into the following sections:

  • The Input - All about recent developments in AI

  • The Algorithm - Resources for learning

  • The Output - Our reflection 

Table of Contents

OpenAI launched the GPT Store on January 10, a platform for custom ChatGPT versions by partners and the community. It offers various Generative Pre-trained Transformer models, especially for ChatGPT Plus, Team, and Enterprise subscribers. The store includes a leaderboard, categories, and a revenue program for creators based on user engagement with their models. (source)

Project Digits: Supercomputing Power for AI Developers

NVIDIA introduced Project Digits, a compact yet powerful desktop AI machine just like any other desktop

  • Unmatched Performance: Equipped with the Grace Blackwell Superchip, Project Digits can run large language models (LLMs) with up to 200 billion parameters—all without relying on cloud infrastructure.

  • Tech Specs:

    • 128GB memory pool

    • Up to 4TB flash storage

    • Can link two units for running 405-billion-parameter models

    • Compatible with Linux-based DGX OS and seamlessly integrates with Windows/Mac PCs.

  • Accessibility and Pricing: Priced at $3,000, Project Digits democratizes access to high-performance AI development, targeting data scientists, researchers, and advanced AI enthusiasts.

Would you like to get hands-on with one of these? How excited are you about the next generation of AI desktops? Let us know in the comments. (source)

Project Cosmos: Redefining Physical AI Training

The Cosmos platform is NVIDIA’s latest solution for training autonomous systems, from self-driving cars to advanced robotics, with unparalleled efficiency.

  • Photorealistic Video Generation: Cosmos eliminates the need for costly physical data collection, simulating real-world environments for training AI systems.

  • Impact on Development: This platform accelerates the training of autonomous vehicles and robotics, ensuring faster deployment and improved accuracy.

GeForce RTX 50 Series: The Future of Gaming Graphics

NVIDIA unveiled the highly anticipated GeForce RTX 50 Series GPUs, including the flagship RTX 5090, setting a new benchmark for gaming performance.

  • Breakthrough Architecture: Powered by the new Blackwell architecture, the RTX 50 Series delivers up to 4,000 TOPS and features 92 billion transistors.

  • DLSS 4 for AI-Powered Boosts: The latest DLSS 4 technology leverages AI to generate additional frames, offering up to an 8x performance boost.

  • Enhanced Gaming Realism: Innovations like RTX Neural Shaders and RTX Mega Geometry provide lifelike textures, animations, and real-time rendering enhancements.

  • Pricing and Availability:

    • RTX 5090 launches in late January 2025 at $1,999.

    • Mid-tier models will follow in February.

With these GPUs, NVIDIA reaffirms its dominance in the gaming industry, delivering unmatched performance and visual fidelity. (source)

Other Key Announcements by Nvidia

  • Thor Robotics Processor: A versatile chip integrating advanced driver-assistance systems (ADAS) with infotainment, designed for next-gen autonomous vehicles.

  • AI Foundation Models for RTX PCs: Pre-built models for creating digital humans, podcasts, and videos, enhancing AI interactions on RTX platforms.

  • DRIVE Hyperion AV Platform: Powered by the AGX Thor SoC, this platform supports safe and intelligent autonomous vehicle development, with partnerships including Toyota and Mercedes-Benz.

  • Automotive Partnerships: NVIDIA is collaborating with major automakers like Jaguar Land Rover and Volvo to integrate its DRIVE Hyperion platform into future vehicles.

The AI Living Room: Samsung's Vision for Intelligent Entertainment

In a captivating reveal at CES 2025, Samsung has painted a picture of how AI will transform our daily viewing experiences from passive consumption to intelligent interaction. Their Vision AI technology isn't just another smart TV feature – it's a glimpse into how our living rooms might become more responsive, understanding, and personally attuned to each family member, including our four-legged friends.

The concept breaks new ground in four key areas:

  • Click to Search transforms casual curiosity into instant knowledge.

  • Live Translate breaks down global content barriers.

  • Generative Wallpaper turns screens into personalized art spaces.

  • Samsung Food bridges entertainment and daily life, connecting what we watch with what we eat.

What's particularly fascinating is how Vision AI weaves together multiple touchpoints of daily life. Imagine watching a foreign film and getting real-time translations, spotting an interesting dish, and having your smart fridge not only provide the recipe but also help you order missing ingredients – all without leaving your couch.

The technology aspires to understand not just what's playing, but who's watching. It's a shift from content-centric to viewer-centric entertainment, where the TV adapts to individual preferences, learning and evolving with the family's interests over time.

This vision raises intriguing questions about the future of home entertainment. How will this level of personalization affect family viewing habits? What does it mean when our TV knows our preferences better than we do? And perhaps most curiously – can AI really understand what makes our pets' tails wag?

As we edge closer to this AI-enhanced viewing future, Samsung's Vision AI represents more than just technological advancement – it's a reimagining of how we might interact with our home entertainment systems in increasingly natural and personalized ways. (source)

Sam Altman maintains a blog which is a treasure trove of thought-provoking insights, blending reflections on technology, entrepreneurship, and the future of humanity. Known for his candid and visionary writing, Sam delves into topics ranging from the potential of AI to societal challenges in adapting to exponential change. 

Recently, Sam Altman published a blog titled "Reflections", commemorating the two-year anniversary of ChatGPT. In this post, he candidly discusses OpenAI's journey toward artificial general intelligence (AGI), highlighting both the achievements and challenges faced along the way.

Some excerpts from "Reflections”-

Launch of ChatGPT: In the winter of 2022, a quiet research lab introduced ChatGPT, marking a significant technological tipping point.

OpenAI's Journey: Founded nine years ago with the goal of developing beneficial AGI, OpenAI's path reflects the broader AI revolution, featuring breakthroughs and governance challenges.

User Growth: Post-launch, ChatGPT experienced rapid growth from 100 million to 300 million weekly active users, showcasing unprecedented AI adoption.

Leadership Crisis: Altman's unexpected dismissal and reinstatement was described as "a dream gone wrong," leading to improved governance and focus.

Non-linear Progress: OpenAI's story involves iterations, learnings, and adaptations rather than a straightforward trajectory.

Future Goals: OpenAI aims for superintelligence, anticipating AI agents in the workforce by 2025, supported by a methodical approach to deployment and safety research.

What’s more interesting is that this blog post coincides with a recent cryptic six-word message he shared on X: "Near the singularity; unclear which side." This phrase encapsulates a critical moment for humanity regarding AI, reflecting both its vast potential and inherent risks as society navigates this transformative landscape. (source)

Microsoft’s Phi-4 Model is Now Open for All

Microsoft has made waves in the AI community by releasing its Phi-4 model as an open-source project on Hugging Face. This 14-billion-parameter model, which excels in reasoning and language tasks, is now available for anyone to download and customize under a permissive MIT License. 

Initially launched on Microsoft’s Azure AI Foundry, Phi-4 has garnered attention for outperforming larger models like GPT-4o Mini in benchmarks such as mathematical reasoning.

With Phi-4's streamlined architecture, developers can now leverage its capabilities without hefty computational costs. This move reflects a significant shift towards efficiency in AI, challenging the notion that bigger is always better. (source)

In a fascinating development at the intersection of AI and mathematical reasoning, a groundbreaking competition is being organized in Cambridge, MA, following o3's recent performance on FrontierMath.

With over $10,000 in prizes and recognition in the FrontierMath publication, it aims to benchmark human mathematical excellence and evaluate AI reasoning, bridging traditional math and AI innovation. (source)

Elon Musk's xAI has completed pretraining Grok 3, marking a significant leap in AI development with 10 times more computational power than its predecessor. The model was trained on the massive Colossus supercluster, featuring approximately 100,000 NVIDIA H100 GPUs.

The scale is staggering:

Current: ~100,000 H100 GPUs

Next phase: 200,000 GPUs

Future vision: 1 million GPUs

As AI companies race toward more powerful models, computational capacity has become the new measure of capability. xAI's strategy of leveraging X's user-generated content for training data showcases how social media platforms are becoming crucial grounds for AI development. (source)

Tim Brooks, a notable former OpenAI researcher, is now spearheading Google DeepMind's ambitious push into physical world simulation. This move represents more than just a leadership change – it's a deliberate stride toward AI systems that can understand and interact with real-world physics.

Building on DeepMind's impressive foundation (Gemini, Veo, and Genie), Brooks brings his crucial OpenAI experience to pioneer AI systems capable of real-time physical world interaction. His expertise bridges the gap between abstract AI models and tangible reality simulation. (source)

Chinese startup NeuroAccess has achieved a groundbreaking milestone in brain-computer interface (BCI) technology, successfully decoding thoughts and speech in real-time. Two trials demonstrated the capabilities of its flexible, high-throughput BCI device, which allowed a patient with a brain injury to control digital systems and enabled real-time decoding of Chinese speech.

In August 2024, neurosurgeons at Huashan Hospital implanted the 256-channel BCI device in a 21-year-old epilepsy patient. The device used high-gamma brain signals to train neural networks, achieving latency under 60 milliseconds and mapping brain functions within minutes.

In December 2024, a second trial focused on speech decoding, with a patient achieving 71% accuracy in synthesizing Chinese speech using 142 common syllables. The system processed single-character decoding in under 100 milliseconds, marking significant progress in connecting cognitive functions with language synthesis. (source)

ByteDance, the parent company of TikTok, has launched INFP, an AI system that animates still portrait photos to make them appear as if they are speaking and reacting to audio. INFP automatically assigns speaking and listening roles, creating realistic conversational videos. 

It utilizes a two-step process: analyzing facial expressions from real conversations and matching audio with natural movements. The system is trained on a dataset of over 200 hours of dialogue, enhancing its performance in lip-syncing and emotional expression. 

Future expansions may include support for text and images, while access remains limited to research institutions to prevent misuse. With platforms like TikTok and CapCut, ByteDance is well-positioned to integrate INFP into its broader AI strategy, further enhancing creative and interactive content. (source)

  • In the latest episode of Leading with Data, we kick off 2025 with an inspiring conversation featuring Navin Dhananjaya, Chief Solutions Officer at Merkle. With over 27 years of experience in data, analytics, and technology for Fortune 1000 companies, Navin shares his transformative journey, including his pivotal role in developing Merkle GenCX and leveraging generative AI to redefine customer engagement. We explore Navin's career-defining moments, his insights into solving complex problems like Amazon’s category classification, and the game-changing potential of AI agents.

  • With AI agents dominating the spotlight, now’s the perfect time to dive into the future of AI. This course, Anyone can Build AI Agents, equips you with the skills to master this transformative field and stay ahead in the AI revolution.

  • Also, check out NVIDIA CEO Jensen Huang’s keynote at CES 2025, where he unveils the groundbreaking RTX 50 series GPUs powered by Blackwell architecture. Discover revolutionary advancements in AI, accelerated computing, and industrial digitalization set to transform industries worldwide!

  • Here’s Bloomberg’s exclusive 2025 interview with Sam Altman, where he delves into the future of AI, his vision for OpenAI, and the profound societal changes ahead. Don’t miss this fascinating read from one of tech’s most influential leaders!

CES 2025 has clearly shifted its focus to AI advancements in the consumer space. What excites you most about these updates? And where do you see the future of AI-powered PCs heading? 

Let us know in the comments!

See you next week!

Reply

or to participate.