@realzny@webuildscore you can use any model that ends with “winner” here: https://t.co/zKbuarwn7u and make your own recipe, they’re all open-source. the nano segmentation model is not.
you’re asking the wrong question.
there is no hidden reward magic. https://t.co/AiTZtuuptC lists Crime, Road Signs, Person as separate bounties with their own data. miners optimized those models. we could then build nano models from or use them separately that’s the entire generalization story.
@LamidaGlobal@mountaindoge100@MaxScore Buybacks will be automatic and visible onchain once Score Studio starts. All revenues (-costs) will go to buybacks and burn.
Score Studio should go live in septembre I guess
training objective is a dual gate: accuracy and size.
new models clear both well enough that one @nvidia jetson handles 35–40 cctv cameras and 7–14 agents.
Those @nvidia Jetsons going into fuel stations just got a lot denser.
With our new models, one unit now runs 35–40 cameras + 7–14 agents. Heavier agents eat more of the budget. Lighter ones scale further.
The cameras were already on site. The intelligence wasn’t.
The newest GTA 6 gameplay trailer dropped yesterday. we ran our subnet vision models on it.
Person, car, gun, segmented and tracked across every scene. night, water, motion blur, freefall over the city. footage the model had never seen.
The part that matters is that this isn't a huge model. it's tiny, and it runs near real time.
Vision that works in the real world...and the fake one.
.@tao_minersunion was the first validator to set Root weights after the Root Reborn update today.
@taoswap_org is the first interface to make them visible.
You can now see exactly how MUV allocates its Root weights directly on Taoswap.
We’ve onboarded several cracked computer vision engineers as alpha testers for Score Studio.
One of them sent this video back with this note:
“Crossed arms are basically an adversarial attack and it still called it. person 0.99, dog 0.98, 42ms / if it can catch a dog mid-dance, it can catch whatever's happening in your warehouse.”
Btw, proving the app's utility by shitposting memes we've run through it could be a great marketing idea
someone smart could create an evals marketplace just filled with subnet validation mechanisms
centralised companies are currently raising banana rounds for their eval marketplaces