Max

Why should I care if a robot gets better? Honestly, I don't. Not because robots aren't important. It's just hard to care about another “robot learned a new task” demo when it has nothing to do with everyday life. What made Axis interesting to me is the problem underneath those demos. A robot can look impressive when everything goes exactly as expected. Put it in a slightly different situation and suddenly that same robot doesn't know what to do. That's where the training loop matters. Axis lets a policy attempt a task, exposes where it struggles, brings in human correction, and turns that experience into data for the next iteration. So the useful part isn't simply having more robot data sitting somewhere. It's learning which failures are actually worth learning from. Axis has reported one real-robot experiment where this loop pushed task success from around 22% to 52%. It has also found that a filtered group of trajectories can outperform throwing the entire data pool into training. That second part is the one I keep coming back to. Imagine having 1 million examples, but most of them teach the robot something it already knows. Now imagine finding the 10,000 examples that expose the things it still gets wrong. Those 10,000 could be worth more than the other 990,000. And that matters because reliable robots are a very different thing from impressive robots. The former can eventually do useful work around us. The latter makes a good demo. That's the bet I'm watching with @axisrobotics : not just collecting more experience, but building a loop where the robot's own failures help determine what it should learn next. Better robots don't start with more hype. They start with better ways of learning from being wrong. Exploring here: s.kaito.ai/XDIwbfa @KaitoAI

What if the Hazels NFT is not actually the final product? That is the part of @0xhazels I find more interesting. The 5,555 characters are being developed as IP that can move beyond the NFT itself — into physical collectibles, holder editions, staker variants, meetups, merchandise and eventually broader RWA products. There is even a defined revenue structure around that physical/IP layer: 5% of positive distributable net product revenue is allocated to the Hazels RWA Pool, with active stakers receiving a higher participation weight than holders. So the NFT becomes the identity of the IP, not necessarily the product people stop at. That creates a different path: character → community → product → physical IP And that is where Hazels starts to look less like a collection with utilities and more like an IP trying to build an ecosystem around itself. The interesting question is no longer how much utility can be attached to a 5,555 NFT collection. It is how far the characters can travel outside the wallet. @0xhazels