Chemofiend

Apple has acquired another AI firm, though this time Ascend.io is aimed at company data management rather than any consumer-facing product. By @WGallagher appleinsider.com/articles/26…

LETS VAMP PUMPFUN START PAIRING EVERY PUMP COIN TO THIS LETS FLIP THEM

@nvidia why? nvidia.com/en-us/startups/

My project to put an Nvidia Jetson inside the Reachy Mini is progressing! One part left to re-print and I'll replace the base with a jetson brain. I'm excited! So is the "customer" in the armchair. @ClementDelangue this feels timely given you all are joining forces with Nvidia. The reachy is great!!

Her: I wonder if he’s thinking about me? Me: Cat szn + MOODENG = Meow Deng

A young man in Thailand reunited with his missing dog while purchasing drinks by the roadside

It feels like a TikTok for humanoid robots. RoboTok, from Rice University and NVIDIA, tackles a practical problem: how to find useful human demonstrations from the huge amount of video online. Instead of searching by captions or objects, it uses 3D hand motion to retrieve videos with similar actions, which can then help humanoid robots learn manipulation tasks like turning bottle caps, screwing faucets, sliding levers, reorienting objects, and bimanual handovers. The interesting part is that this improves downstream robot learning. On three challenging dexterous tasks, RoboTok improved success rates by 17.8, 38.0, and 59.8 percentage points over the strongest baselines. --- Humanoid robots still lack enough manipulation data. The industry is trying to distill the physical world into data, while also mining the internet for useful traces of human interaction. The data gold mine for Physical AI is constantly moving. How much of this data can ultimately translate into more capable humanoid labor in the real world will be a long-term challenge, and perhaps the real benchmark.