@0xTheGreatApe@taoleeh you just don’t get any rewards if you spot an object on an image that is not showing any objects from the class you were supposed to spot
building the validation mechanism for this was my favourite part. it will start with a generative engine, where miners will be scored on footage they've never seen or touched. later, it will move to real live streams, and miners will be scored against what happens next in the real world.
@MaxSebti Curious how validators will score a world model. Next-frame prediction against real footage, or downstream tasks like spotting events in a scene?
SN44 Score already pays ~34 $TAO/day in emissions across 256 neurons, so the miner incentive is there.
spooky season
this is the output of our first world model that will be trained by our miners
llms understood text and ate software. world models will understand the physical world and eat operations.
billions of cameras are already installed. almost none of them understand what they're seeing.
most vision ai treats detection and segmentation as the end game. "there's a truck." so what?
for us they're just the eyes and sensors. the world model is the brain. it knows the delivery is late, the bay is blocked and who needs to know.
detection alone means retraining for every new client. that's a services company, not a product. a world model learns how the physical world works once, then applies it everywhere.
and it makes the eyes sharper. a world model can generate endless realistic footage of any site, any angle, any weather, any rare event you'd wait years to catch on camera. fires, break-ins, accidents, on demand. detectors trained on that see things no real dataset could ever show them.
eyes to see, a brain to understand, act and keep training the eyes.
every company will need this to make their operations autonomous. we're building it.
Kimi K3.1 Pro is on Nous Portal, so I gave it the Mont-Saint-Michel voxel prompt from my series, with one hour of /goal in Hermes Agent.
it built the abbey with its golden statue, the village inside its ramparts, pilgrims on the causeway, sheep in the salt meadows, a tide coming in, a rain shower and a night with lit windows.
it used the whole hour. its last edit to the page came at 44 minutes, and it spent the rest testing the page in a browser and checking its own screenshots.
the run cost $3.38.
If you Moneyball your startup idea, you end up selling data
If you follow your emotional startup idea, you end up building a personal assistant
If you build something that people want, you end up forward deploying AI into car dealerships in middle America
One of the things I’ve been testing at @webuildscore. Can’t say much yet, but you can see the difference.
On top, the boxes jump around and miss half of what happens between frames. Below, they just stay locked on.
It’s a small thing until you’ve watched the choppy version for a few weeks. More soon.
normal cv engineers are building another toyota
my delusional team is building a new tesla
beyond the car analogy, the real challenge is not to build another bounding box machine
it is to automate physical businesses
"company brain" startups that construct information from internal messages and documents when they realize 90% of information only exist in peoples heads: 😭😭😵💫😵💫
China's Unitree just dropped H2-X Kung Fu Legend Guardian.
Dents car doors. 4 meter drops. Spinning kicks + flips. Even lifts front of a car.
We are cooked 💀