In partnership with
🌟 Editor's Note: Recapping the AI landscape from 12/09/25 - 12/15/25.
🎇✅ Welcoming Thoughts
Welcome to the 23rd edition of NoahonAI.
What’s included: company moves, a weekly winner, AI industry impacts, practical use cases, and more.
Started today’s newsletter in Chicago and finished up in Cleveland.
Highlighted a TED Talk in today’s interview section, very interesting, somewhat polarizing company.
Lots of new reports releasing as the year comes to a close.
Interesting report on the State of Enterprise AI from OpenAI last week.
Cool story on Google Co-Founder Sergey Brin coming out of retirement to save the company.
Another interesting report on the whirlwind of a year at Meta.
Two months ago this week from OpenAI would’ve been a slam dunk win.
Lots of different topics post-race this week.
The more I think about today’s Startup Spotlight the more I like the concept.
Meta and xAI are back in their usual positions.
Let’s get started—plenty to cover this week.
👑 This Week’s Winner: NVIDIA
NVIDIA is back at the top. Not necessarily a “wow” week for them, but the consistency and speed continues to impress. Here are their highlights:
Acquired SchedMD: NVIDIA acquired SchedMD, the company behind the open-source workload scheduler Slurm, widely used in HPC and AI. Sometimes these headlines sound made up. Essentially this is another vertical acquisition aimed to manage compute traffic in data centers.
Unveiled Nemotron 3 Models: NVIDIA announced the Nemotron 3 family of open source AI models, designed for agentic AI tasks. The smallest model, Nemotron 3 Nano, was released immediately. Basically handing out reference models for building AI agents that run best on NVIDIA hardware. Just strengthening their position here.
GPU Geolocation: NVIDIA created software to track and verify the location of its GPUs to ensure restricted hardware does not end up in unauthorized jurisdictions (China). Doing what they can to serve all markets.
NVIDIA seems to be executing at a high level everywhere they can. More details are also emerging around the China situation. NVIDIA’s top-tier GPU (Blackwell) remains off-limits, H200 GPU sales will now come with a 25% cut for the U.S. government, and at the same time China may be favoring domestically developed alternatives, but there are conflicting reports. Nothing is definitive yet, and I’ll continue to monitor how this plays out.
From Top to Bottom: Open AI, Google Gemini, xAI, Meta AI, Anthropic, NVIDIA.
⬇️ The Rest of the Field
Who’s moving, who’s stalling, and who’s climbing: Ordered by production this week.
🟣 Google // Gemini
Browse to Build: Google launched Disco, an experimental Gemini-powered browser feature that turns open tabs into on-the-fly mini apps (GenTabs) for research, planning, and learning. Very cool. Excited to try this out.
Deep Research API: Gemini Deep Research was rebuilt on Gemini 3 Pro, adding an API so developers can embed the research agent. Great! Been waiting for a deep research API for a while. Hallucination goes way down with context and DR provides that.
Gemini-Translate: Google upgraded Translate with Gemini, adding smarter text translation, live speech-to-speech translation, and expanded language learning features. Nice.
🟢 OpenAI // ChatGPT
GPT-5.2 Launch: OpenAI released GPT-5.2 with Instant, Thinking, and Pro tiers, but early reactions were mixed, with some users calling it bland or less creative than earlier models. I don’t see a huge difference from 5.1. Don’t need a splash release but would like to see more improvement over time.
Disney $1B Partnership: Disney is investing $1 billion in OpenAI and licensing 200+ characters for Sora, bringing powerhouses such as Marvel, Star Wars, and Pixar into AI video generation. Big move here on the consumer and content front, I like it.
Adult Mode Confirmed: OpenAI plans an optional, age-verified “adult mode” in Q1 2026, pending stronger age-check systems. Eh. Makes sense strategy wise.
🟠 Anthropic // Claude
Accenture Partnership: Anthropic partnered with Accenture to train ~30,000 consultants on Claude and Claude Code, accelerating large-scale enterprise AI deployments in regulated industries. Comes right after the OpenAI x Accenture partnership. Good enterprise move.
Safety First: Anthropic is pushing open, safety-focused agent standards by donating MCP to the Linux Foundation and co-founding the Agentic AI Foundation with OpenAI. I didn’t realize Anthropic created MCP. Cool. They have stayed consistent in valuing safety.
AGI Risk Warning: Chief Scientist Jared Kaplan warned that allowing AI to self-train by 2027–2030 could trigger an uncontrollable intelligence jump. AGI is something we’ve talked about for a while. I think the benefits outweigh the perceived risk
🔵 Meta // Meta AI
Bye-Bye Open Source: Meta is moving away from open-source LLaMA toward a proprietary model, Avocado, expected in early 2026, emphasizing tighter control and monetization. Probably the right move to keep competitive.
China Ad Revenue Concerns: New studies found ~19% of Meta’s China ad revenue came from scams and banned content after AI moderation and anti-fraud teams were scaled back. Not ideal. Do they care?
AI-Powered Ad Tools: Meta rolled out AI tools to help brands find creator content and quickly turn it to partnership ads, aiming to boost ad performance and scale creator marketing. May kill a few startups but this is a really smart move.
🔴 xAI // Grok
El Salvador Initiative: xAI partnered with El Salvador to deploy Grok across 5,000+ public schools, bringing AI-powered tutoring to over 1 million students. Good move here.
Bondi Beach Misinformation: Grok spread incorrect details about the Bondi Beach shooting, highlighting ongoing risks of AI errors during breaking news. This is going to happen more with xAI because of its social media placement. Irregardless, they need to get it fixed.
Tesla Optimus Update: Tesla shared new Optimus footage featuring Grok and showing smoother movement and object handling, signaling continued momentum toward real-world humanoid robots. Excited for Tesla’s progress with humanoids and other robotics.
🤖 Impact Industries 🏥
Robotics // Cultural Restoration
Researchers have developed an AI-powered robot that can reassemble fragile Pompeii artifacts by scanning fragments, predicting how pieces fit together, and physically reconstructing objects with minimal human handling. Using computer vision and probabilistic matching, the system reduces the risk of damage while accelerating restoration work that would normally take months or years. It shows how embodied AI can preserve cultural heritage, not just automate modern industry.
Medical // Home Monitoring
Researchers propose a low-cost, audio-only AI system that monitors people with dementia at home using microphones instead of cameras or wearables. The system listens for speech, emotional cues, and everyday sounds like coughs, cane taps, or thuds, then generates daily caregiver reports highlighting confusion, agitation, safety risks, and health changes. Designed for low-income or remote households, it aims to scale caregiving support without replacing human judgment.
💻 Interview Highlight: Palmer Luckey → AI in Defense
Interview Outline: Palmer Luckey presents a warning about a possible "Second Pearl Harbor" scenario against China, arguing that the U.S. will fail due to its lack of capacity and outdated, slowly-produced technology. The solution he proposes is the rapid adoption of mass-produced, software-defined autonomous systems powered by AI, which he argues is both a necessity for deterrence and a more ethical path for warfare.
About the Interviewee: Palmer Luckey is the founder of Anduril Industries and Oculus VR. After pioneering consumer virtual reality, he turned his focus to defense, building AI-powered autonomous systems designed to modernize U.S. and allied military capabilities.
Interesting Quote: ”Your Roomba has better autonomy than most of the Pentagon’s weapons systems.”
My Thoughts: Palmer Luckey is someone I’ve been following for a while now. He’s a bit of a mad man, but is very efficient and intelligent in what he builds. I’m no expert in AI defense tools, but what Palmer discusses here makes sense. Autonomous and connected AI robots on the front lines seems the future of battle. The quote from below: “Refusing AI doesn’t make war safer” is a good one and can be restated many ways across different industries.
Featured Interview Highlight — Palmer Luckey on AI & Modern Defense
Source: TED Talk — “The AI Arsenal That Could Stop World War III”
Q: What’s fundamentally broken in today’s U.S. defense strategy?
Luckey: The problem isn’t spending — it’s speed and capacity. The defense industry moves slowly, while Silicon Valley has largely disengaged. As a result, consumer AI systems often outperform far more expensive Pentagon technology.
Q: Why is the phrase “killer robots” misleading?
Luckey: Autonomous weapons aren’t new — landmines and naval mines already operate without humans. The real choice is between smart systems that can distinguish civilians from threats, or dumb weapons that can’t. Refusing AI doesn’t make war safer.
Q: How does AI help counter China’s massive numerical advantage?
Luckey: China can outbuild the U.S. by orders of magnitude. We can’t win by matching numbers. AI enables scalable force multipliers — autonomous systems that can be produced quickly and aren’t constrained by human manpower.
Q: Why must autonomous systems operate without constant human control?
Luckey: If a weapon depends on a remote signal, it can be disabled by jamming. A system that fails when communications drop is a liability. True autonomy is required for systems to function in contested environments.
Q: What is Anduril’s core contribution to modern warfare?
Luckey: Anduril’s Lattice platform is software that coordinates large numbers of autonomous systems. It allows defenses to improve at the speed of code — updating faster than adversaries can adapt.
👨💻 Practical Use Case: NotebookLM
Difficulty: Basic
NotebookLM is Google’s AI research notebook built to work only with the sources you provide. Instead of pulling from the open internet, it reads your documents, links, PDFs, and transcripts, then helps you summarize, analyze, and ask questions based strictly on that material. Think of it less like a chatbot and more like a personal research assistant. It’s a very similar concept to project folders in GPT and Claude.
NotebookLM is best when you’re dealing with information-heavy work, such as:
Summarizing long articles, PDFs, or research papers
Breaking down interview transcripts or meeting notes
Creating briefs, outlines, or memos from source material
I provided it with the 2025 Mckinsey AI report from a few weeks back so I could dive in a bit deeper, here’s what it looks like:
On the flip side, from a business perspective, say you’re preparing for a meeting with a potential new client. You can upload:
Baseline company details
Relevant articles or reports
Your Internal SOP’s for context
Within minutes, NotebookLM can outline next steps, answer questions pulled straight from the documents, and help you prepare talking points before you ever get on the call. I like the ability to look at data in a bundle of different ways, which is very helpful if you learn best using methods other than simply reading.
Learn more about it here ⬇️
👃🏼 Startup Spotlight
Osmo
Osmo - Teaching AI Sense of Smell
The Problem: AI can see, hear, and speak, but it still can’t smell. That’s a massive gap in how machines understand the world. Scent is deeply tied to human memory, emotion, food, and health, yet it’s largely absent from AI models. Without smell, we’re missing a critical sensory dimension for industries from perfumery to diagnostics.
The Solution: Osmo is building the world’s first AI models that can interpret and generate scent. Using machine learning, chemistry, and vast scent data sets, Osmo’s tools can predict how molecules will smell and even design new fragrances. Their goal is to decode the language of smell, enabling applications in flavor science, synthetic fragrance, and even disease detection through scent biomarkers.
The Backstory: Osmo was spun out of Google Research in 2023 and is led by Alex Wiltschko, a neuroscientist who previously spearheaded Google’s AI smell research. The company’s name nods to “osmosis” — blending biology, chemistry, and computation. Backed by Lux Capital and others, Osmo is turning a once-impossible idea into a commercial frontier.
My Thoughts: When I first came across this, I kind of thought it was a joke. But, looking further into it, the concept makes a tone of sense. Our ability to smell is often second nature, yet very useful. Digitizing Smell is a baseline that can lead to preventative care (think dog or cat understanding a human is sick before we can), amongst countless other concepts, namely robotics-based over time. Seems obvious in retrospect.
“It’s not likely you’ll lose a job to AI. You’re going to lose the job to somebody who uses AI”
- Jensen Huang | NVIDIA CEO
If you want to learn a new word every day, just search up NVIDIA headlines. Till Next Time,
Noah on AI
Find customers on Roku this holiday season
Now through the end of the year is prime streaming time on Roku, with viewers spending 3.5 hours each day streaming content and shopping online. Roku Ads Manager simplifies campaign setup, lets you segment audiences, and provides real-time reporting. And, you can test creative variants and run shoppable ads to drive purchases directly on-screen.
Bonus: we’re gifting you $5K in ad credits when you spend your first $5K on Roku Ads Manager. Just sign up and use code GET5K. Terms apply.