There’s a privacy problem hiding inside one of the most basic questions in crypto “prove you have the money” if you answer onchain, you can point to your wallet but now the other side can see the wallet too @primus_labs is approaching this differently with zkTLS instead of revealing the underlying financial data, you can prove a specific threshold was met the flow is pretty straightforward → log into the exchange or custodian through a normal HTTPS session → an attestor verifies that the session is authentic → no login credentials or raw account data are exposed → you receive a signed attestation saying something like “balance above X” → a smart contract verifies that attestation onchain the counterparty gets the information they actually need, without automatically getting the information they don't that's a subtle but important change in how financial verification can work and it's already being applied to proof of reserves for Unitas USDu their dashboard combines attested offchain hedge balances with onchain positions the catch is that this isn't permanent proof an attestation represents one verified snapshot, so anything that needs to stay true has to be re-attested over time Join Primus here: s.kaito.ai/imYMgiu

Applied for my Identity See you in the @RYFTCORP ryft.fun/whitelist/r-3b16fc6…

I think @FungoLabs is onto something that NFTs have been missing for years. The problem with most NFTs is that the moment something interesting exists, the entire internet can see it. Rare trait? Scraped Hidden rank? Indexed Special ability? Someone already knows There’s basically no room for an actual secret Fungo is approaching this differently with encrypted NFTs on Ethereum, powered by Zama’s FHE. The NFT can stay encrypted on-chain, and the holder decides when the underlying essence gets revealed. Even the team can’t see what’s hidden That’s the part I find genuinely interesting This isn’t just putting a privacy label on an NFT It gives digital ownership another dimension: information you control because you own the asset Seen by everyone Known by one I like that idea Applications open → fungolabs.org/apply

One detail in the @axisrobotics docs deserves more attention than I initially gave it every submitted trajectory receives a unique data ID and is associated with both the exact task and the contributor the record is then anchored onchain on Base that's important because it turns provenance into part of the data architecture rather than a separate metadata layer so the reported 2.1 million trajectories aren't simply an aggregate count in principle, individual records can be traced back to the contributor and task that produced them that creates a much stronger foundation for understanding where training data actually comes from especially for robotics, where the quality and context of a trajectory can matter just as much as the trajectory itself most AI datasets don't give you this level of visibility into the origin of every sample you get a collection of data, but the chain of custody is often unclear Axis is taking the opposite approach the provenance is created at the moment the data enters the system that's a small architectural decision that could become a very important one as the dataset scales Join Axis here: s.kaito.ai/69icJQQ

If you wanna know @opencampus_xyz how works, read this:

Been seeing more people talk about minds.games lately, and I finally started paying attention. There’s something about the MDG ecosystem that feels like it’s still early enough to be interesting. Not trying to force a narrative here, but I definitely want to see what they’re building before everyone starts talking about it. who else is keeping an eye on MDG? 👀

At some point, saying “we need more robot data” stops being the interesting part @axisrobotics may be getting close to that point 100,000+ trajectories generated every day means the system is no longer operating like a traditional data collection project it's becoming a continuous production engine humans generate experience simulation expands it models create feedback the next round gets better and when that loop runs at this scale, the real challenge moves somewhere else how do you separate useful signal from noise? how do you turn it into training data for VLA and world models? and most importantly, how do you make what one robot learns useful for another embodiment? that's where Axis's work around selective DAgger, simulation-real co-training and shared datasets becomes important because the end goal isn't having a huge folder full of trajectories it's building an infrastructure where robotic knowledge can keep being generated, refined and transferred Physical AI becomes much more interesting when data stops being something you occasionally collect and starts becoming something the system continuously produces Join with me here: s.kaito.ai/69icJQQ

There’s an interesting opportunity developing around YT-onyc on @ExponentFinance right now The base APY is roughly 11%, compared with an implied APY of around 12.9%, and that spread makes the current pricing worth looking at A $1K $ONyc position can produce around 214K @onrefinance points daily while generating close to $8.46 in yield each day If held until maturity, that’s roughly $839 in total yield and about 21M points After subtracting the yield generated, the effective cost for those points is only $161 I think this is a good example of why point markets can become interesting when the underlying yield is doing most of the heavy lifting keep watching @onrefinance closely as one of the more interesting DeFi point opportunities this year Start your Onre journey here: app.onre.finance/earn/leader… Also create YT-ONyc position here: app.exponent.finance/en/mark…

the bigger story behind @jumperapp isn’t really the token it’s the consumer layer Jumper is trying to build around it crypto still makes users jump between bridges, exchanges, chains and different financial apps Jumper’s bet is that this experience can eventually collapse into one interface and now $JUMP gives that ecosystem its own value-accrual layer if Jumper can keep expanding the amount of onchain activity users can do without leaving the app, the super-app thesis gets stronger bridging was the entry point the destination is much bigger Make the JUMP

starting my @primus_labs journey today their campaign just went live on @KaitoAI and Season 1 runs for 30 days you can publish up to 10 tweets, but only your best 4 will count 0.3% of the total supply is allocated to rewards: 40% → top 150 English creators 40% → Chinese creators 20% → Korean creators time to see what Primus is building and put some real thoughts on the timeline join early here: s.kaito.ai/imYMgiu

The latest @axisrobotics work looks less like traditional robot learning and more like an attempt to build a software stack for physical skills the foundation is an expanding Axis Library rather than a giant undifferentiated dataset individual capabilities can be reused, composed into longer-horizon behaviors and distilled into lighter policies their recent experiments give some useful signals expert models reach near 100% success for roughly $5–10 of compute per task, while chained capabilities can handle more complex behaviors a proxy model then evaluates trajectories based on utility instead of treating every sample equally that filtering step matters because a smaller, higher-quality subset can outperform the complete data pool then there's embodied RSI, where the deployed policy itself produces real-world data that becomes training material for a stronger successor the reported improvement from 22% to 52% shows what that feedback loop can look like the benchmarking layer is just as important Open Axis Benchmarking with openroboto continuously refreshes the task distribution using the growing Axis Library so models can't simply optimize against one frozen benchmark they have to keep generalizing as the library evolves that's a pretty different architecture for robotics collect skills → compose them → deploy → learn from failures → benchmark again it's much closer to continuous integration for physical intelligence than simply collecting more trajectories Join here: s.kaito.ai/69icJQQ