🧠 Nvidia enters the AI model race

Good morning! NVIDIA is quietly becoming the open-source king. It just dropped an open 30B model that outperforms GPT-OSS and Qwen3-30B, while running 2.2–3.3× faster. Also, I’ll show you how to turn any document, website, or video into infographics and slide decks.

In today’s AI newsletter:

• Nvidia Launches Nemotron 3 Open AI Models

• Zoom’s Federated AI Beats Gemini on Humanity’s Last Exam

• DeepMind Trains Robots Using Veo World Models

• How to turn any data into infographics and slide decks

• 4 new AI tools worth trying

AI MODELS

🧠 Nvidia Launches Nemotron 3 Open AI Models

NVIDIA introduced Nemotron 3, a new open model family designed for agentic AI systems where multiple agents collaborate, reason long-term, and route tasks efficiently. The first release, Nano, prioritizes speed, efficiency, and transparency over raw scale.

• Supports a 1M-token context window for reasoning over long documents, codebases, and logs

• Uses a hybrid MoE architecture with far fewer active parameters to reduce costs

• Delivers up to 4× higher token throughput than Nemotron 2 Nano

• Fully open: model weights, datasets, and RL tools released on Hugging Face and GitHub

As AI moves from chatbots to autonomous agents, efficiency and long-context reasoning become critical. Nemotron 3 signals NVIDIA’s bet that open, cost-effective models will power real-world agent systems at scale, not just demos.

AI NEWS

📊 Zoom’s Federated AI Beats Gemini on Humanity’s Last Exam

Zoom revealed that its federated AI system scored 48.1% on Humanity’s Last Exam, a tough expert-level reasoning benchmark, surpassing Google Gemini and placing it among the top-performing AI systems.

• Federated system orchestrates models from OpenAI, Anthropic, Google, plus

• Zoom’s own models via a “Z-scorer”

• Score beats Gemini 3 Pro (45.8%), but trails GPT-5.2 (50%)

• Powers AI Companion 3.0, improving summaries, reasoning, and task automation

• Claim is disputed by DeepWriter, which reports a 50.91% score

Zoom is betting that model orchestration beats single-model dominance. If federated AI proves more reliable in real workflows, this could signal a shift in how enterprise AI systems are built, less about the best model, more about the best coordinator.

AI & ROBOTICS

🤖 DeepMind Trains Robots Using Veo World Models

DeepMind released a new paper where its Veo video generation model is fine-tuned on robotics data and used as a world simulator to evaluate robot policies on ALOHA 2 bimanual robots, no physical trials required.

• Veo predicts real-world robot success with 0.88 correlation across 1,600+ real-world trials

• Generalizes to out-of-distribution scenarios like new objects, backgrounds, and environments

• Uses action-conditioned video generation with multi-view consistency for realistic simulations

• Enables testing rare and dangerous cases (collisions, failures) safely before deployment

This is a major step toward training and validating robots entirely in simulation. If scalable, Veo-style world models could massively speed up robotics development, cut costs, and make generalist AI agents far safer before they ever touch the real world.

HOW TO AI

🗂️ How to Turn Any Document, Website, or Video into Infographics and Slide Decks

In this guide, you’ll learn how to use Google’s NotebookLM to transform long documents, articles, and YouTube videos into clean infographics and professional slide decks in minutes.

🧰 Who is This For

• Founders and operators preparing decks quickly

• Students and researchers summarizing complex material

• Creators turning long content into visuals

• Anyone who wants clear explanations from dense information

STEP 1: Access NotebookLM

NotebookLM is Google’s AI research assistant powered by Gemini models. It becomes an expert on your documents and links instead of pulling random internet data.

Open your browser and go to Notebooklm. Sign in using your Google account (Gmail or Workspace).

On the homepage, click New Notebook. This creates a dedicated workspace where all your sources, notes, and generated visuals will live for this project.

STEP 2: Add Your Source Material

This step gives NotebookLM the information it will analyze and turn into visuals.

Inside your notebook, find the Sources panel on the left and click Add source.

You can add content in multiple ways:

• Upload files from your computer (PDFs, Google Docs, text files)

• Link files directly from Google Drive

• Paste a website or article URL

• Paste a YouTube video link (NotebookLM automatically processes the transcript)

Once added, NotebookLM starts analyzing the content. This can take anywhere from a few seconds to a couple of minutes, depending on length—for example, a long PDF or a 30-minute video.

STEP 3: Generate an Infographic

The Infographic tool turns dense material into a single, easy-to-understand visual summary.

On the right side of the screen, open the Studio panel. This is where all generative tools live, including Infographic, Slide Deck, and Audio Overview.

Click Infographic. You can optionally click the pencil icon to adjust settings like detail level or layout (square, portrait, or landscape).

NotebookLM generates the infographic in the background. Once it’s ready, you’ll see a clear visual breakdown of the key ideas and how they connect.

Use the menu options to download the infographic as a PNG file and share it anywhere.

STEP 4: Generate a Professional Slide Deck

To create a full presentation, stay inside the Studio panel and click Slide Deck.

Before generating, you can customize the output by clicking the pencil icon. Options include:

• Detailed deck for reading or emailing

• Presenter slides with clean visuals and key talking points

You can also add a custom instruction like:
“Create a slide deck for executive leaders with a focus on ROI and strategic impact.”

Once generated, you’ll have a complete, structured presentation. You can view it in slideshow mode or download it as a PDF to upload into Google Slides or PowerPoint for final edits.

Nvidia acquires SchedMD, the developer of Slurm, an open-source AI workload management system, and says it will keep distributing Slurm on an open-source basis.

Allen Institute for AI launches Bolmo 7B and Bolmo 1B, claiming they are “the first fully open byte-level language models”, built on its Olmo 3 models.

Google is shutting down its dark web monitoring tool, launched in 2024 to alert users when their personal info is detected on the dark web, on February 16, 2026.

AI drug discovery startup Chai Discovery raised a $130M Series B led by Oak HC/FT and General Catalyst at a $1.3B valuation, taking its total funding to $225M.

👀 Cursor: Visual editor with drag-and-drop + AI agents

🪩 Disco: Google’s experimental browser that builds web apps instantly

🧠 GPT-5.2: OpenAI’s faster, smarter model with improved reasoning

📚 Gemini Deep Research Agent: AI that handles long research tasks

THAT’S IT FOR TODAY

Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!

See you tomorrow :)

- Dr. Alvaro Cintas