gm everyone💙 whats your plan for today?

Putting financial activity onchain creates a real privacy problem. institutions can’t expose every position, strategy, balance or customer detail just to make their activity verifiable. @primus_labs is working on this part of the stack: ▸ zkTLS verifies Web2 data without exposing the source data ▸ FHE allows computation while data stays encrypted ▸ zkFHE adds verifiable computation with privacy ▸ TEE helps protect sensitive execution [straightforward: -: public settlement. -: Private financial logic. for onchain finance to work with sensitive financial data, both sides matter. exploring Primus XP: s.kaito.ai/ZltPSz7

gm everyone💙 whats your plan for today?

One thing I keep thinking about with @axisrobotics is that Physical AI isn’t just a model problem. It’s an experience problem. Robots need huge amounts of real world interaction to understand how to move, adapt, recover from mistakes, and complete tasks reliably. That’s where Axis is building something interesting. → 5M+ robot trajectories on Base → 200K+ contributors → Open Axis Benchmark → Franka dataset → Research accepted at IROS 2026 The bigger idea is the feedback loop: Human interaction → Data → Training → Evaluation → Better performance → Better robots. As AI moves from screens into the physical world, real world experience could become one of the most valuable resources. The companies that can build the strongest data flywheel may ultimately have a huge advantage in Physical AI.

The next big robotics company might not build a single robot. It might build the infrastructure that teaches millions of robots how to act. That’s the thesis I’m watching with @axisrobotics. One of the biggest bottlenecks in Physical AI isn’t just building better robot hardware. It’s getting enough high quality, real world training data. Axis is building around that problem with a full data loop: → Create tasks in simulation → Put humans in the loop → Capture robot trajectories at scale → Verify and structure the data → Train models → Identify failures → Generate new tasks → Feed the results back into training And the scale is already interesting: • 5.48M+ verified trajectories • 7,700+ verified tasks • 53K+ hours of trajectory data • 196K+ contributors Why does this matter? LLMs had the internet to learn from. Robots need experience. They need data on grasping, picking, placing, navigating, manipulating objects, and recovering when something goes wrong. That creates a different kind of scaling loop: MORE TASKS → MORE DATA → BETTER MODELS → BETTER ROBOT CAPABILITY → MORE USEFUL DATA The real question isn’t simply whether robotics becomes a huge industry. It’s whether the infrastructure that continuously generates and improves robot experience becomes a critical layer of the Physical AI stack. That’s the part of @axisrobotics I’m paying attention to. The robotics race isn’t only a hardware race. It’s a data race. And we’re still early. 🔗 s.kaito.ai/9kNz0ol

the @axisrobotics pre-sale is coming to an end, but honestly, that’s not the part I’m most interested in. what matters more is what comes after. Axis already has people completing tasks, generating robot data, and contributing to a growing community around Physical AI. now imagine that system scaling: «more contributors more tasks more real-world data more feedback for training robots» that’s where things get interesting. the pre-sale can bring attention to Axis, but the real test is what the team builds with that momentum afterward. the next phase is where the network has to prove its value.

A lot can be said about a token sale, but one thing deserves to be recognized: @axisrobotics is listening to the community and making adjustments where it matters. The new bonus mechanism is a meaningful move toward rewarding the people who believed in Axis early, even under strict unlock conditions. Here’s what changed: 🔹 25% of tokens unlock for everyone at TGE 🔹 Bonus tokens are fully unlocked at TGE, based on the original commitment before pro rata dilution 🔹 Fulfillment rate increased from 41.7674% to 42.0346% after suspicious wallets and KYC submissions were rejected 🔹 Refunds have already been sent directly to participants’ wallets no claim needed With 2,472 participants, this wasn’t just about raising funds. It was about identifying genuine supporters, protecting the integrity of the sale, and giving early believers recognition for their conviction. Not every decision will satisfy everyone, but adjusting the mechanism while keeping the focus on genuine participants and long-term Physical AI development is something worth acknowledging. Early supporters believed in the vision. Now Axis is showing that conviction matters. Excited to see what comes next. @axisrobotics

Today my sorsa score huge up by +19 Total score: 120 Whats your sorsa score??

one thing about onchain data still feels strange to me: we want everything to be verifiable, but that doesn’t mean everyone should see the data behind it. that’s what makes @primus_labs interesting. Primus is building infrastructure for private data verification and computation using zkTLS + FHE. the idea is simple but powerful: prove that something is real and valid without exposing everything behind the proof. i’m digging deeper into Primus and will be sharing what i learn along the way.

Clean demos can teach a robot what success looks like. But real world tasks are rarely clean. When a policy reaches a state it hasn’t seen before, that’s where things can break down. @axisrobotics turns those failures into part of the learning loop: Task → Policy tries → Failure → Human correction → New data → Better policy The key isn’t avoiding failure. It’s learning from exactly where the robot fails. That’s what makes the training loop more adaptive and ultimately helps robots handle the real world better.

The more I explore @axisrobotics, the more I think Physical AI has a deeper problem than simply needing more data. The real question is: what data should we collect next? A trajectory can be perfectly valid and still add very little if the model already understands that behavior. That’s where the loop gets interesting: collect → train → evaluate → identify the gap → collect again Instead of treating failures as dead ends, Axis can use them to understand where the model is still struggling and guide the next round of data collection. So the question shifts from: “Is this trajectory good?” to: “Does this trajectory teach the model something it doesn’t already know?” Human intervention becomes useful here too. Rather than repeating complete demonstrations, people can step in at specific failure points and turn those corrections into targeted training data. And with 5M+ robot trajectories coming from 200K+ contributors, the challenge is no longer just dataset size. It’s making every new contribution more informative. Epoch 2 of the Creator Program is already live. Epoch 1 ended with 5,341 creators, 250K+ creator link clicks and 31,980 signed trajectories referred. The interesting part now is seeing which contributions actually help close the next gap in robot learning.

Privacy is magic 🪄 Join now : ShieldedWizards.com

JOIN THE HUNT. Apply now: apply.bitfoots.xyz/

gm @IThinkItsArt family💙 whats your plan for today?

The deeper I go into @axisrobotics, the more I understand why the Base integration matters. Physical AI isn’t just about building better models. It needs massive amounts of real-world data—and that data needs a clear history of where it came from and who helped create it. That’s where the onchain layer becomes interesting. Robots generate the data through real interactions. Base can provide a transparent record around those contributions, making provenance and ownership easier to track. So the blockchain isn’t replacing the robot training process. It adds a coordination and ownership layer around the data powering it. That’s the part of Axis I find particularly interesting.