Nylig så jeg at @0xhazels begynner å bli nevnt ganske mye, så jeg prøvde å finne ut av det. Hazels er en samling på 5 555 NFT-er, født ut av historien om Eight Rise. Det er verdt å merke seg at prosjektet ikke bare fokuserer på NFT-er, men bygger et Attention Layer som knytter seg til skapernes og fellesskapets aktiviteter. For øyeblikket har Hazels Studio GTD Race med 500 plasser, i tillegg til puslespillbrikker som Pulse og Drops. Hazels' idé er ganske klar: å knytte NFT-eierskap til oppmerksomhet, bidrag og tilgang i stedet for bare å eie en PFP. Det er fortsatt ganske tidlig, så vi må følge med på hvordan dette systemet faktisk fungerer. All informasjon undersøkes av deg selv, du bør gjøre din egen undersøkelse før du deltar eller tar noen beslutninger. @0xhazels #WEAREHAZELS

Robots Are Missing Something AI Has Had From The Start There’s a fundamental difference between AI and robotics that often gets overlooked. AI has the internet. For decades, humans have written books, papers, documentation, code, conversations, and countless other forms of content. Without necessarily intending to, we created a massive dataset for machines to learn from. When humans describe something, they create data about it. Robots don’t have that luxury. Humans have been cooking, opening doors, pulling drawers, folding clothes, and manipulating objects for thousands of years. But most of those interactions never became datasets. The knowledge of how humans interact with the physical world is still distributed across billions of hands rather than sitting somewhere a robot can learn from. And this is where I think Axis becomes interesting. Instead of waiting for millions of robots to be deployed before collecting enough data, Axis turns humans into part of the data engine. No robot required. No robotics lab required. No specialized hardware required. Through a browser, contributors can generate trajectories that teach robots how to interact with the physical world. But there’s an important paradox. The more open the system becomes, the harder data quality becomes to control. If you only reward volume, you eventually get farming, bots, and trajectories created purely for rewards. That’s why verification, replay, quality scoring, and anti-bot mechanisms are much more important than simply watching the trajectory counter go up. Because the thesis behind crowdsourced robotics isn’t: “More data is always better.” It’s: “More diverse ways of interacting with the world give robots a better chance to understand that world.” One person opens a drawer one way. 10,000 people might open it in 10,000 different ways. Some will be precise. Some will be messy. Some will be fast. Some will be slow. But that variation creates coverage. And that’s the part I find most interesting about @axisrobotics A contribution doesn’t have to end with Points. It can become part of a dataset, feed into training, improve a policy, and eventually contribute to systems operating in the real world. Axis Dataset V1 already showed a glimpse of this: continual pretraining lifted π0.5 from 83.9% to 88.8% on LIBERO-Plus. At the same time, the contributor network is expanding rapidly, surpassing 200K registered users in August. That’s the bigger picture. Axis isn’t simply trying to collect millions of trajectories. It’s experimenting with something much larger: a global data network for robots, where millions of people each contribute a small piece of the physical world. AI had the internet as a natural data layer. Perhaps Physical AI needs its own version of that for the physical world. And Axis is betting that: the internet of robots could be built by humans. #axisrobotis

RT @baovy06: After following @axisrobotics for a while, one change that stands out to me is the launch of the Axis Challenger Program. It…