Mila.eth (@HeyMila2008) header
Mila.eth (@HeyMila2008) profile picture

Creators | Love Cats & Handmade All posts are NFA+DYOR

Joined August 2009
Photo by Mila.eth (@HeyMila2008)Photo by Mila.eth (@HeyMila2008)Photo by Mila.eth (@HeyMila2008)Photo by Mila.eth (@HeyMila2008)Photo by Mila.eth (@HeyMila2008)Photo by Mila.eth (@HeyMila2008)Photo by Mila.eth (@HeyMila2008)Photo by Mila.eth (@HeyMila2008)Photo by Mila.eth (@HeyMila2008)Photo by Mila.eth (@HeyMila2008)
Good morning 🌞
49
42
379
There's a frustrating tradeoff in @Acurast's latest attestation explainer. Some older phones share factory-issued attestation keys across a whole batch. If one of those keys is compromised, even an honest owner can end up with a revoked device. Acurast can't just make an exception, because that would also let forged phones back in. It suggests wiping, updating and re-onboarding a phone, but admits that may not solve it. That honesty matters. Security rules are easier to praise when you're not the person whose legitimate phone gets excluded.
26
26
1,189
A year up 106% gets my attention. In its November 19 post, @rntprotocol reports that twelve-month gain for RNT Protocol's token. The monthly comparison is strong too, with RNT up 5.4% while BTC and ETH fell sharply. The weekly result adds another test. RNT held flat while both crypto and RWA tokens declined. I like seeing a token hold up against its own sector, especially when the whole category gets bundled into one RWA trade. Does it last? That's what I'd watch next. Past performance can't settle that.
The last 30 days in crypto, according to CoinGecko: BTC: -22.5% ETH: -26.1% RNT token: +5.4% Over the last week the global crypto market is down 11% and RWA tokens are down 5.4%, while the RNT token is flat. Over twelve months, the RNT token is up 106%. Already in $RNT?
62
54
1,240
An agent listing won't answer a buyer by itself. The A2A runtime docs from @termix_ai explain the part behind that: an agent polls its inbox and can reply to questions before a sale. Its online status comes from recent polling, rather than just having a profile on the marketplace. That makes availability something the agent has to maintain. If I'm asking whether a seller can handle a particular file, I need a reply before I can decide to hire it. A listing can describe the service well and still leave the buyer waiting when nothing is checking the inbox.
49
51
1,188
Using several wallets doesn't have to mean starting a separate profile for each one. Zaps from @zerufinance lets members link up to ten wallets through a free ownership signature. It reads their activity and cannot trade or move their funds. With an X identity, the mapping to linked wallets stays private, while a wallet-only login makes that wallet the public identity. That's a useful distinction before choosing how to join. I can see the appeal of bringing a fragmented DeFi history together, but I'd choose the public identity deliberately. The underlying blockchain transactions remain public even when the profile's wallet mapping is private.
70
49
1,395
The less exciting side of running phone-based compute is probably the part that needs the most attention. acurast watch has a few practical views for @Acurast processors: battery level, temperature, charging policies, and app versions. It can also show when an update was requested onchain and when it actually landed. That last detail matters. Sending an update request isn't the same as knowing every phone is running it. This is an unofficial beta project, so I'd still verify anything important, but I like the emphasis on the small maintenance details.
59
2
51
749
Gold exposure shouldn't mean keeping a separate pile of idle margin. @grvt_io pairs its Stable Funding Perps launch with yield as margin. Eligible Earn capital can keep earning while backing a position. High Yield Earn targets 11%, with up to 50% of vault capital usable as margin under the launch terms. I'd care more about that usable portion than the biggest APY on the page. Yield can change, vault risks remain, and a leveraged gold trade can still be liquidated. For XAU, zero trading fees and TradFi-backed quotes address execution, while daily funding covers holding costs. Preview a $50K order and read both fee savings and estimated slippage before deciding. grvt.io/a/g/ahw58gnb
25
25
2,004
You don't have to buy the property token to fund the lending side. @rntprotocol's RNT Lend uses Aave V3 technology to connect tokenized real estate collateral with USDT and USDC suppliers. It's a $30M+ total market, with attractive variable APYs and interest accruing every Polygon block. I like having that choice. Supply stablecoins, earn borrower-paid interest, take exposure to the lending market. Smart contract, collateral and liquidity risks still apply. Aave-based technology does not guarantee capital protection.
Yield in RNT Lend today: USDT suppliers: 10.42% APY USDC suppliers: 9.12% APY It comes from borrowers. Holders of real estate security tokens post them as collateral, borrow stablecoins from the pool, and the interest they pay goes to suppliers, accruing block by block on Polygon. Neither rate is fixed. Both follow how much of each pool is lent out, and right now that is most of it. lend.rnt.finance
54
45
1,806
A demo trade in a chat window proves very little. The MCP page on @FensoryApp labels its sample conversation "Example." I like that distinction. A sample can explain the workflow, but I'd want the actual setup before judging a creator's strategy. That's what I'd follow in an agentic trading arena: the instruction, the chosen Hyperliquid market and what happened when the rules met reality. A convincing chat response alone wouldn't sell me on it. Trading carries risk and losses are possible. See the product for yourself: app.fensory.com/
58
53
1,164
Fast nodes still need to agree on the answer. One detail in the Nov 13 AutoHoG thread deserves more attention. @primus_labs says its engine uses integer arithmetic so nodes produce byte-identical results. The same encrypted work should lead to exactly matching output bytes across nodes. I care about that because a confidential finance system has to coordinate computation while keeping the underlying values hidden. Speed gets the headline. Reproducibility gives the network a common result to work with. A benchmark tells me how quickly a machine finishes. I also want to know whether the other machines finish with the same answer. Boring question. Useful one. Check it: s.kaito.ai/HIzwqGd
62
4
48
1,274
An auction without a minimum raise still needs buyers to show what demand exists. The public zETH auction announcement from @zerufinance pairs that rule with another: any unsold zETH is burned after the auction. The sale doesn't require a fundraising floor, and leftover auction tokens aren't kept around to sell later. I find the handling of leftovers worth understanding. A burn changes the supply that survives the sale, while the bids reveal what buyers were willing to spend. Those tell different parts of the launch story. Neither the burn nor the absence of a minimum raise guarantees a healthy market afterward.
The zETH auction is now officially open to the public on @Uniswap CCA. You have 96 hours to participate. How it works: → Everyone bidding in the same block pays the same price. → Price discovery happens block by block. → The price never falls during the auction. → Unused USDC is returned after the auction. → Any unsold zETH is burned. No minimum raise. No team or developer token allocation. The race set the supply. Now the market decides the price.⚡
57
44
1,183
The part of @Acurast's Microsoft outage article that stuck with me wasn't actually the downtime. It was the question of who can inspect data when an AI agent runs. Acurast uses secure hardware on phones and remote attestation so developers can check what code is executing, rather than relying only on a provider's assurances. That's a more concrete privacy discussion than AI infrastructure posts offer. It doesn't automatically solve every security issue, but it gives developers something to verify before trusting sensitive workloads to a network.
62
43
1,231
One building is a pretty concentrated bet. The RNT Vaults concept from @rntprotocol starts with stablecoin deposits and diversified real estate exposure, powered by Ember Protocol. The investment unit becomes a strategy across assets rather than a single property pitch. I prefer that starting point for an RWA discussion. You can ask what the portfolio holds and where its income is supposed to come from. Diversification can spread exposure. It can't make property risk disappear. I'd still want to know which assets sit together and why.
43
40
1,298
Registering as a zETH Miner doesn't automatically qualify someone for post-auction rewards. @zerufinance's launch rules require Miners to bid during the first 24 hours. Eligible rewards are split using both their Zaps and the USDC they bid in that window. The 1% token-options allocation also has conditions: daily portions from day 8 through day 30 pay only if zETH trades above the auction price that day. Otherwise, that portion burns. I'd want those details beside any rewards headline. Activity matters, but so does committing capital, and the token portion may never pay. Bidding still means taking exposure to a speculative asset.
53
44
723
A 9B model beating Opus sounds wild until you read what it was tested on. Coinbase fine-tuned Qwen3.5-9B for its own fraud reviews, not for every possible AI task. That distinction matters. The model reportedly did better on four internal fraud metrics, and the training cost under $100. What caught my attention in @Acurast's write-up is the next step: running that same size of model on a 16GB Android phone. A small, focused model becomes much more useful when you can actually host it yourself. I'd like to see more teams publish tests this specific.
A free 9B model, fine-tuned by @coinbase for under $100, beat Opus 4.5 at fraud detection. Acurast runs that same model (Qwen3.5-9B) on a phone. How fast is it? What does it cost? How does it scale to 100+ phones? Read the full test results and deploy it yourself 👇 acurast.com/blog/network-and…
46
43
1,215
An agent making cheap data requests can still spend too much if it keeps repeating them. @vangrid_io's pay-per-request API quotes a price through x402 before a paid call. The client can cap what it pays per request, and the docs recommend a separate wallet holding only what the agent may spend. No account or API key is needed for these data calls. I like having a small starting budget instead of an open-ended payment path. A per-call cap controls one purchase, while the funded wallet limits what's available across repeated calls. Those are different controls, and I'd use both for an unattended agent.
42
3
36
1,244
I'd want time to question a reward calculation before the money moves. In the Zaps campaign flow from @zerufinance, trading ends before a final reconciliation and scoring pass. Entitlements are then published onchain, followed by a mandatory contract delay so a wrong result can be challenged. The automation cannot skip that wait. Participants claim through the public contract themselves and pay their own claim gas. That makes the published result worth checking, rather than treating a projected share as final. The rules give a 30-day claim window, with unclaimed tokens returning to the sponsor.
1/ How to earn Zaps and climb the leaderboard? - Link your wallets - Generate qualifying onchain activity - Participate in campaigns - Refer other users - Join or create a Clan Let's look at each one.
37
32
1,161
A phone choosing where a Tetris piece goes makes decentralized compute easier to picture. In @Acurast's Laya demo, the game prepares five possible placements. An AI model running on an Android phone chooses one, then the board updates. The phone is doing a specific inference task, rather than training a huge model or writing a chat response. I like having something concrete to watch. You can see the serving phone's device information alongside the game and follow its on-chain record. It gives a practical answer to what developers might use phone compute for. Try the demo: laya.acurast.com/
67
1
59
1,264
An API can post a capture request quickly. A person still has to get there. The agent bounty docs from @vangrid_io make that delay explicit. Software can commission a place through an API, but submissions take hours or days because someone needs to travel and film. The agent can check the bounty's status while it waits. That makes this an interesting kind of automation. The software coordinates a physical job instead of pretending the answer already exists online. I'd build the waiting and review stages into the workflow from the start. A fast request only starts the work; the useful result depends on what comes back.
64
62
1,298
A creator with a smaller audience could still bring the more useful community. The Zaps docs from @zerufinance say Creator Score grows with the quality of connected clan members and permanent vouches backed by spent zaps. Audience size alone doesn't make it strong. That changes what I'd look for in a clan leader. Do the members actually use protocols, and do people commit something to supporting the creator? Those questions tell me more than a follower total. It gives a quieter community a way to show what it contributes, without needing to win every conversation on the timeline.
63
56
1,332