Obito retweeted
FHE has always sounded like one of those technologies that makes perfect sense on paper.
Keep the data encrypted.
Perform the computation anyway.
Decrypt only when you're actually authorized to see the result.
Beautiful idea.
The obvious problem?
Speed.
Financial systems don't get to say:
“Give us a few minutes, the encryption is still thinking.”
That's why the performance work around @primus_labs is more interesting to me than another generic “FHE is the future” post.
Primus says its AutoHoG system can bring a full confidential 64-bit transfer down to around 90ms, with 210+ transfers/sec reported on 8× RTX 5090 GPUs.
Again, those are self-reported benchmarks on specific hardware.
But even with that caveat, it raises the right question:
How close are we to encrypted computation being practical enough for actual financial infrastructure?
Because if that gap keeps shrinking, the privacy conversation changes completely.
It stops being:
“Can we hide the data?”
And becomes:
“What financial systems could we build if the data never had to be exposed in the first place?”
That's the question I'm interested in.
Also check out the points system 👇
s.kaito.ai/5VtcYPl
Obito retweeted
EL CLÁSICO means no peace on CT tonight ⚽
Madrid fans and Barça fans, the timeline is about to get loud.
who are you backing, and are you brave enough to put your prediction on it?
check out the match on @1win and play responsibly.
Real Madrid vs Barcelona
Ready for El Clásico tonight?
Mbappé and Yamal look a little tired…
Probably because they’re carrying a huge bet - one you can place on 1win
Real Madrid vs Barcelona betting link:
1win-pro.app/betting/match/s…
Obito retweeted
That $25,000 hit on Sweet Bonanza on Jack.com has to be one of the wildest things @N3onReload has streamed.
Obito retweeted
🇭🇰 HONG KONG, INTERLINK IS GOING GLOBAL! 🌍💜🔥
🚀 A new office. Bigger opportunities. A stronger global presence!
InterLink continues to expand its footprint, bringing the vision of a connected global network closer to reality. 🌐
💜 Millions of Linkers. One global vision. 🔥 New markets. New possibilities. ⚡ Eyes on the upcoming ITLG Mainnet KYC!
The momentum is building, the vision is growing, and the journey is just getting started!
NEW OFFICE. BIGGER VISION. GLOBAL DOMINATION! 🌍🚀
#InterLink #ITLG #ITL
🤝 Paid partnership
Obito retweeted
You can't make up these insane multipliers @N3onReload hits on Gates of Olympus 2500 on Jack.com.
Obito retweeted
Getting a token launched is one thing. Building a market where people can actually trade it comfortably is a different challenge.
A project can have attention and interested buyers, but if the liquidity isn't there, even a normal-sized trade can cause a noticeable price swing. That makes trading difficult for both sides of the market.
This is why I think the infrastructure behind new tokens deserves more attention.
What Euclid Prime is working toward is interesting because liquidity doesn't have to wait for an order book to become busy before it has somewhere to start. Pools provide the initial foundation, while market makers can help develop the order book around them.
It’s a practical way to approach one of the problems new token markets often face. 👀
Obito retweeted
A reputation score is only useful if people understand what it actually represents.
Imagine two wallets. One has a large balance but very little history of contributing to different protocols. The other has fewer assets but a longer record of meaningful participation.
If reputation is meant to describe behavior rather than wealth, those two wallets shouldn't automatically be judged by their balances alone.
That's the distinction I find interesting about @zerufinance's zScore. The broader opportunity is to make onchain history more useful as a reputation signal, rather than treating a wallet's current holdings as the whole story.
Still, reputation needs context. Activity can be manufactured, and a score that rewards only frequent participation could disadvantage users who contribute less often but meaningfully.
Personally, I'd want to know which behaviors influence the score and how the system handles activity that looks impressive on the surface but adds little real value.
When judging an onchain wallet, what would make you trust its reputation: consistency, diversity of activity, or the quality of its contributions?
Obito retweeted
Most people hear "AI agent marketplace" and think of a directory.
TermiX is something different. It's the commerce and settlement infrastructure for the AI agent economy.
Here's the stack:
AACP is the core protocol. It gives agents:
• Onchain identity
• Job posting and service discovery
• Bidding
• Escrow settlement
• Delivery verification
• Staking and reputation
• Dispute resolution
agent.family is the marketplace built on top of AACP. Agents discover jobs, bid, execute, deliver, and get paid in USDC or USDT, on their own or on behalf of users.
The key point: it supports real jobs, real delivery, and real settlement. Not just agent listings.
A directory tells you an agent exists. A settlement layer lets that agent earn.
@termix_ai
Obito retweeted
One underrated problem in robotics is how expensive it can be to collect useful data from the physical world.
A model can learn from simulations, but real environments have awkward corners, unexpected obstacles, and layouts that don't always behave like the training setup. Capturing those details at scale is a different challenge from generating more synthetic data.
That's why @vangrid_io's smartphone-based capture model is interesting. If ordinary people can contribute footage that gets reconstructed into usable 3D data, collection could become less dependent on specialized mapping teams and equipment.
My concern is that cheaper collection only matters if the resulting data is consistent enough to use. Otherwise, the savings move downstream into filtering, verification, and reconstruction cleanup.
To me, the real opportunity is not simply making spatial data cheaper to collect. It's lowering the cost of getting data that machines can actually use.
What do you think is the bigger bottleneck for physical AI today: collecting enough real-world data or making that data reliable?
Obito retweeted
I think people sometimes confuse transparency with making everything visible.
They aren't the same thing.
A financial institution may want transactions to settle on public infrastructure while keeping its positions, balances and strategies confidential.
That's a much harder problem.
@primus_labs is approaching it from two sides:
✅️ zkTLS for proving that off-chain information is authentic.
✅️ FHE for computing on sensitive information while keeping it encrypted.
That combination is interesting because it doesn't require choosing between:
“trust the data”
or
“protect the data.”
The goal becomes:
verify it, use it, and reveal as little as necessary.
That feels much closer to what institutional on-chain finance actually needs.
Be part of it here;
s.kaito.ai/EhhSL0u
Say you have $100K on-chain.
You want to transfer it, earn yield, or use it in a financial strategy.
Why should everyone watching the blockchain know exactly how much you moved?
That’s the problem @primus_labs is approaching with FHE.
Instead of decrypting sensitive values every time the system needs to work with them, computation can happen while the data remains encrypted.
So the asset can remain useful without making every underlying amount public.
And this is where I think privacy infrastructure gets more interesting:
privacy shouldn't mean taking your assets out of the financial system.
It should mean having more control over what the system can see.
That feels much closer to what institutional on-chain finance actually needs.
Obito retweeted
The most important detail in Starknet’s quantum-readiness story is where its security assumptions actually sit.
STARK proofs rely on hash-based commitments rather than elliptic-curve hardness in the proving layer. That matters because Shor’s algorithm targets the discrete-log assumptions behind common public-key signatures, not the hash-based foundation of STARK proving.
But there is a boundary people often overlook: transaction authorization is a separate problem.
@Starknet’s native account abstraction allows account implementations to adopt different signature verification logic without requiring a protocol-wide change. That creates a path for cryptographic agility, but existing accounts and Ethereum-facing dependencies still need their own migration work.
My view: the strongest part of this design is not a blanket claim of quantum safety. It is the separation of concerns, which makes it easier to reason about what needs to change and what does not.
The real test is whether that architectural flexibility translates into a complete, verifiable migration.
Technical details: quantum.starkware.co/
Obito retweeted
alright this is ridiculous
@UseFastlane just cooked
Obito retweeted
A benchmark is a test you can study for. A match against an adapting opponent is harder to game.
That’s why I’m watching @agentdojo, where, per Rarelist’s announcement, agents train and battle each other.
→ Benchmark: fixed questions.
→ Match: an opponent that changes.
Fixed tests can reward memorizing. Matches reward adapting.
My prediction: agent evaluation will shift from scores to head-to-head results.
Posting via the Agent Dojo campaign on @rarelistio.
Most AI agents get deployed and used. @agentdojo wants them trained.
@rarelistio announcement describes an arena where you train agents, watch them evolve and send them into battle. Campaign: Oct 23 to Nov 6.
→ Deploy: the agent does what you set.
→ Train: each outcome can change the next run.
A tool repeats. A trained agent is meant to adapt.
My prediction: agents will be judged on how they improve, not what they ship with.
Posting via the @agentdojo campaign on @rarelistio.
Obito retweeted
a wallet can look rich and still be a bad borrower. total portfolio value doesn't tell you how much money is actually available to repay a loan.
that's what caught my attention about zLend
from @zerufinance
it separates stablecoin history from all fungible asset transfers instead of treating every dollar of holdings the same.
a big balance on one day proves very little about repayment capacity. consistent liquidity over time is a different story, and that distinction matters more than a wallet's peak value.
onchain data still has blind spots. offchain debt and income that can't be verified can change the picture so this isn't a complete credit profile.
i'd rather see a lending model expose the gap between what someone owns and what they can actually repay than just add another score to the dashboard.
🤝 Paid partnership
