@404gen_ SN Already with product producing 3D Models. 10M Mcap
@Corcel_X SN19 LLM online with Users. 20M MCap
@rayon_labs SN64 & 56 has Users. 94M Mcap & 38M Mcap
@webuildscore SN44 soon online product with partners. 9M Mcap
Buying what has a product & great team backing it $TAO
@etienne_breton Michael burry still doesnt understand the tech even tho there's literally hundreds of companies using it and dozens of government branches using it
@StockSavvyShay Jesus it takes a few news articles to drop the market. And a whole earnings report to bring it back up.
The market is so fearful and irrational
@DrocksAlex2@markjeffrey Llms are easily copied if they give up their weights. But to trully help a company with Llm you would have to give them up.
Alex is saying that there needs to exist a third party/layer that llms use, that is built specifically for each business. That's what palantir does
tamashi controlling a robot?
can a small model actually picture the world, or is it just guessing?
this week i went looking for the answer inside JEPAs to build what will be our next, next, training run on sn44.
a world model squeezes each camera frame into a small internal vector.
the question is what that vector actually knows. so i fit a simple linear probe from it to the true physical state, tested it on episodes it had never seen, and compared against shuffled labels to rule out luck.
if it beats chance, the model genuinely represents that quantity.
block position came back at R²=0.97, so it clearly knows where things are.
orientation is still weak, so it doesn't yet know how they're turned.
that gap is exactly where the work goes next.