In partnership with
๐ฑ
World models are about to change everything.
AMD just dropped $8 billion on Fei-Fei Li's startup to build AI that understands physics, not just words.
This isn't about chatbots anymoreit' s about machines that can simulate reality itself.
But wait, there's more:
Andrej Karpathy built a YouTube alternative in two hours using Claude
Anthropic just bought $3.4 billion worth of gas generators to power their AI dreams
Iโm Alex. Welcome to L8R by Innov8.
Letโs dive deep.
In todayโs post:
World Models Will Soon Be Inescapable. Here's What You Need to Know
Andrej Karpathy built a curated video platform in a weekend using Claude and 20 years of Hacker News
Anthropic just made the largest known power purchase by any AI lab: $3.4B for 1.7GW of gas generators
AI Is Learning Physics, Not Just Words
Chatbots are old news.
The biggest tech giants are now pouring billions into world models that actually understand physical reality.
AMD just dropped over $8 billion on Fei-Fei Li's startup, World Labs, to dominate this wild new space.
The details:
World models predict the next frame of a physical environment instead of guessing the next word in a sentence.
They use video and sensor data to build interactive 3D worlds where AI learns gravity, motion, and how objects interact.
AMD bought World Labs for over $8 billion to get ready for a massive robotics push expected in 2027.
Heavy hitters like Google DeepMind with Genie, NVIDIA with Cosmos, and Meta are rushing to build their own reality simulators.
These virtual worlds let robots practice dangerous tasks safely before trying them in real life.
Why it matters:
We are moving past AI that just talks to us. We are entering the era of Physical AI.
By 2027, this tech will power the home robots and driverless cars we have been promised for years.
If a robot can master a task in a perfect digital twin of our world, it will not mess up in your living room.
๐ก L8R's Take:
Text models are hitting a wall, and this is the obvious next step.
If you want to build the ultimate AI, it needs to understand how the real world actually works.
AMD spending $8 billion shows they know exactly where the smart money is heading.
From Our Partner
Introducing Fair Game from Vanity Fair
Join Mark Guiducci for a series of in-depth and far-reaching conversations with the most interesting characters from the worlds of Hollywood, Politics, Style, Silicon Valley, and beyond. It's all Fair Game.
Karpathy Built a Video Platform in 2 Hours Using Claude
Andrej Karpathy just proved that building a complex app takes hours, not months, if you know how to use AI.
He built a brand new video platform called HackerTube using Anthropic's Claude.
It is a massive library of the best YouTube videos recommended by smart coders over the last two decades.
The details:
Karpathy used Claude to read 43 million Hacker News comments from 2007 to 2026.
The AI found 6,454 videos that at least three different people recommended over time.
Claude Opus 5.5 acted like the boss, while smaller Sonnet 5.5 agents did the boring work like writing tags and video cards.
Karpathy only did three things himself: getting an API key, renting a server, and buying the domain name.
He launched the site on October 31, 2026, and it runs without annoying algorithms or clickbait thumbnails.
Why it matters:
This shows a massive shift in how we make software.
One person with a smart AI can now build huge, data-heavy apps in a single afternoon.
It also proves people are tired of junk recommendation algorithms and want real human taste back.
๐ก L8R's Take:
Karpathy is showing us the future of coding right here.
The days of giant teams building basic websites are totally dead.
If you are not using AI agents to build your ideas right now, you are just wasting your own time.
From Our Partners
Why Overseas Producers Can't Undercut Atlas Salt, Even for Free
Cities are already running short on road salt, and winter hasn't even started. Some supply bids are going unanswered, prices have nearly doubled in parts of the market, and the backup plan is two weeks away by boat. Here's why Atlas Salt (TSXV: SALT | OTCQX: SALQF) can get salt to the people who need it most, faster and cheaper than anyone shipping from overseas:ย
โข Three days by ship to Boston, not fourteen
โข Shallow deposit that's cheaper and faster to mine
โข Projected C$188M a year in free cash flow
โข Export Development Canada interested in lending C$150M
โข Early supply deal with top U.S. salt distributor
Anthropic Drops $3.4B on Giant Gas Generators
Anthropic just dropped a massive $3.4 billion on gas-fired generators.
They are buying 1.7 gigawatts of power equipment to keep their AI data centers running.
This is the biggest power grab we have ever seen from a single AI lab.
The details:
Anthropic bought the equipment commitments from a Texas supplier named OnePWR Solutions.
The deal includes heavy-duty Caterpillar generators that OnePWR ordered two years ago.
By taking over these old orders, Anthropic skips a massive multi-year wait line for new hardware.
The huge price tag covers both the generators and the engineering needed to set them up.
Why it matters:
AI needs a crazy amount of electricity.
Frontier labs cannot wait for slow utility companies to catch up.
They are taking matters into their own hands.
By buying the hardware directly, Anthropic guarantees their data centers will have the juice they need to train next-gen models.
๐ก L8R's Take:
You cannot build the future of AI if you cannot keep the lights on.
Anthropic is smart to skip the line and buy the heavy metal directly.
If you rely on traditional power grids right now, you are going to lose the AI race.
๐ Quick L8R Summary
World Models: AI labs are racing to build systems that simulate physical reality, with AMD dropping $8B on Fei-Fei Li's World Labs to get ahead.
HackerTube: Andrej Karpathy built a curated video platform in 2 hours using Claude, featuring 20 years of Hacker News recommendations.
Anthropic's Power Play: Anthropic just made the biggest energy purchase by any AI lab $3.4B for gas-fired generators to power their data centers.
๐ฉ Innathe L8R engane undarunnu ๐?
We read every reply - just reply to this email and let us know how we can improve !
Appo adutha L8R il kanaam bieโฆ๐