Shinra 👁️

1.6K posts

Shinra 👁️ banner
Shinra 👁️

Shinra 👁️

@ShinraHQ_

Omnichain retard @onchainhaven ⎸@Watchers_FNF

Ethereum Mainnet Katılım Eylül 2022
900 Takip Edilen1.7K Takipçiler
Sabitlenmiş Tweet
Shinra 👁️
Shinra 👁️@ShinraHQ_·
Threw this together for the homies in the trenches ➩ watzemeta.co It's basically a viewer of all the narratives that we know with the "best" coins on it so you can see which meta is going or not Feel free to test it
English
3
4
18
1.1K
Shinra 👁️ retweetledi
floppa
floppa@gryptocaracal·
thanks gmgn for not allowing me to sell, the best platform fr gmgn.ai/r/floppa
English
8
3
32
1.1K
Shinra 👁️
Shinra 👁️@ShinraHQ_·
@mac_eth There no real reason if i have to be honest with you, coingecko is an incredible tool, but gecko terminal but Habits are hard to break especially in crypto where we keep using the thing we're feeling safe with
English
0
0
0
45
Mac
Mac@mac_eth·
@ShinraHQ_ what's the reason? Maybe GeckoTerminal will fix it
English
1
0
1
156
Shinra 👁️
Shinra 👁️@ShinraHQ_·
@NTmoney @GrailCo Once liquidity issue is solved, it can be great, but having 3.4k LP in a 650k mc token is not gonna attract anyone
English
0
0
0
86
Nick Tomaino
Nick Tomaino@NTmoney·
Mission at @GrailCo is to make it fun and easy to bet on people you believe in. We do it by vaulting real world collectibles and minting tokens that can be redeemed for them. 10,000 gWEMBY = 1 Silver Prizm PSA 10 WEMBY rookie card. Helped build Coinbase, Polymarket, Opensea from early days and applying a lot of lessons learned to Grail. With your help, this can be the next consumer app that onboards millions of new people.
English
19
12
136
8.4K
Base
Base@base·
Hearing that you can't say "Crypto and AI" without instantly thinking about Base
English
222
60
575
22.7K
Shinra 👁️
Shinra 👁️@ShinraHQ_·
I can explain in a more techy way but it's harsh lmaoi On the llm part, looked at the engines SPEC and it actually returns q2 char-mlp C16 E40 H1-256 H2-128 V9, so its a character level MLP not a transformer, more like the kinda archi Karpathy ships in his makemore tutorial, the large model is 21kb of weights and the small one is 128 BYTES, so technically its a tiny next char predictor more than an LLM in the modern sense, doesnt really change the thesis tho, if anything it makes the cultural angle even stronger cause the «immortal model onchain»
English
0
0
0
115
Shinra 👁️
Shinra 👁️@ShinraHQ_·
Small thing I wanna flag, I looked at the actual specs and the "LLM" thing is kinda misleading, it's not a real LLM like ChatGPT it's a much simpler kind of model that predicts text one letter at a time Which actually makes it more interesting to me not less, once you accept the model is a tiny archi the real product is the immutability and the fact it's a factory, anyone can deploy their own weights through it and get 20% of every inscription on their model forever, so it's not one LLM onchain it's a permissionless marketplace of immortal mini models with QUILL as the fuel, way cooler frame than free LLM for everyone tbh There is still some sketchy shit tho: - the deployer wallet is 22h old, anon, no ENS, and they also deployed two other QUILL tokens on the side plus alternate versions of the factory and engine (probably just test runs before mainnet but you gotta be careful nobody apes the wrong contract) - lp locker and vesting are there which is the bare minimum hygiene for a token launch, - no third party audit on the factory tho and that's the one holding peoples QUILL deposits It's basically (but i'm not sure of that) the first immortal char level predictor and somehow that's enough to make it matter ngl
English
1
0
0
113
Shinra 👁️
Shinra 👁️@ShinraHQ_·
If that shit is real, it's criminaly underpriced imo Basically tech is: They got a neural net system running directly inside the chain, using only intergers and they did a smart trick since you only care which output is biggest, they skil a whole expensive math step, they store the model weights as contract code insted of storage making it even cheaper the browser version and the onchain version give exactly the same outputs The core loop is also hand tuned at a level you usually see in GPU kind of shit not smart contracts But there is some black point that i'm trying to figure out, from what i see, neural nets exist since 2022 or smth like that, output are mediocre but the innovation here is the packaging itself not really the science behind it, and it got only improved from here But it's interesting because it's a LLM modele existing only onchain so absolutely nobody is hosting it, if chain still going LLM still going, no server or admins of any kind can change that, so it's free forever and for anyone The tech breakthrough is not that big but the cultural thing behind it is kinda really cool ngl
Quill LLM@quillcomputer

a language model that runs entirely onchain Quill is live on @Base, the model's weights live inside a contract, and every character it writes is computed by the EVM itself no oracle, no server, no off-chain compute you call generate(), the chain does the math, text comes back the $QUILL token is now on Uniswap: 0x60a646e3fd75cde4c5b604b22d4fcd04639913c8

English
2
1
8
432
Shinra 👁️ retweetledi
Dinduz
Dinduz@CryptoDinduz·
Been looking deeper into $QUILL and here's my thoughts: The breakthrough isn’t necessarily “omg they invented new AI architecture.” Neural nets onchain have existed in some form for years. The interesting part is HOW they packaged it. They managed to run a neural network directly inside the chain using integer-only computation, while stripping away unnecessary expensive operations wherever possible. One clever optimization: instead of fully calculating probability outputs, they basically skip the costly normalization step because for inference you often only care about which output score is highest anyway. They also store model weights as contract code instead of regular storage, making execution dramatically cheaper. And apparently the browser inference and onchain inference produce identical outputs, which is actually pretty impressive consistency-wise. The low-level optimization work is also kind of insane for a smart contract project. Some of the core loops are manually tuned in ways you normally associate more with GPU optimization than Solidity. Now realistically: the raw model quality itself still isn’t revolutionary. Outputs are still relatively primitive compared to centralized frontier LLMs. But honestly I think that misses the point. What’s culturally fascinating is the idea of an LLM that exists entirely onchain. No servers. No hosting provider. No API company. No admin keys. No one can silently change the model behavior. As long as the chain exists, the model exists. Forever accessible, forever permissionless, forever reproducible. That’s a pretty powerful concept even if the actual intelligence level is still early. Feels less like “better ChatGPT” and more like the beginning of fully sovereign onchain AI infrastructure. Still only around a ~600k market cap and a relatively new pair though, so obviously be cautious.
Dinduz tweet media
English
3
6
27
6.3K
Shinra 👁️
Shinra 👁️@ShinraHQ_·
@yuzu_computer Of course sir, and you have all the proofs for that? the access mention in your email? You can show us in your sessions that you had one from far away from where you from?
Shinra 👁️ tweet media
English
0
0
1
53
Yuzu
Yuzu@yuzu_computer·
@ShinraHQ_ We were hijacked during the delegation process, and this was unmistakably our error.
English
1
0
0
126
Yuzu
Yuzu@yuzu_computer·
Everyone, please calm down. We launched our app for the very first time today, and even though there's still a lot to refine, why would we launch a token now? Gathering real users and traction, and tokenizing based on that, is the obvious choice, isn't it? We received a DM from someone impersonating a VC, and while verifying a few things in that DM, our account was hijacked. Immediately after we noticed, we recovered it and took the post down. We'll be more careful with account management going forward. We apologize.
Yuzu@yuzu_computer

Don't buy the token, account compromised.

English
6
0
3
936
Yuzu
Yuzu@yuzu_computer·
Don't buy the token, account compromised.
English
26
0
3
1.8K
Mac
Mac@mac_eth·
To the people DMing me about token price, Please instead DM or comment on other people, asking them if they know anyone who can sell Inference on surplusintelligence.ai We sold out. Have buyers lined up. Need more supply.
English
15
2
54
23K
Shinra 👁️ retweetledi
XY
XY@xydotdot·
Why You Never Take Profit You don’t fail because you can’t find winners. You fail because you never close them. Most traders don’t fail at entries, they fail at endings. The exit is tied to a number, and a number has no boundary. As price rises, the brain updates the goal in real time. That update loop is built in: dopamine prefers variance, social proof lifts expectations, and the “goal-gradient” pushes you to stretch the target the closer you get. A number does two things to your psychology. First, it creates a moving threshold: any level can be justified as “almost there,” so decisions are perpetually deferred. Second, it protects ego: raising the target lets you preserve the identity of the brilliant holder a little longer. Both forces bias you toward indefinite postponement. Nothing in your life actually changes at a price on a screen, so the mind finds no reason to stop. Purpose breaks that loop because it introduces a state change your brain can recognize. Paying off a liability, buying time, funding runway, securing an asset....these are discrete outcomes. They flip your environment from one state to another, which gives the brain closure. Closure reduces the tension to keep negotiating with the future. You convert “maybe more” into “definitely mine,” and the decision holds. Purpose also rewires regret. Numbers anchor you to the counterfactual of “what if I held longer,” which is unbounded and will always win the argument. Purpose anchors you to certainty: the debt is gone, the runway exists, the asset is owned. Humans regret lost certainty far less than missed extra upside. That asymmetry is what makes purpose durable under pressure. There’s a second-order effect people miss: profits with purpose improve the next round. Lower stress, fewer forced trades, more patience, clearer time horizons. Capacity goes up, error rate goes down. You didn’t just take money out; you increased the quality of your future decisions. In a game with infinite rounds, that compounding edge matters more than squeezing the last leg of a single move. Numbers drift; purposes hold. Give your wins an outcome that exists outside the market, and the exit stops feeling like betrayal of opportunity. It reads as completion, and it funds a stronger next hand.
English
51
92
628
41K
Grail
Grail@GrailCo·
S/o to our design director the legend @Coldie When you collect GENESIS PACKS you get 3 Grail NFTs created by the man himself You can hold the NFTs or redeem them for gTOKENS which you can hold, redeem for the physical card or sell for cash There are many layers to Grail that we hope you will understand over time. Appreciate the love ❤️
English
9
6
40
3.3K
Shinra 👁️ retweetledi
Max
Max@Maxhyperliquid·
They tried to stop me The job isn't done yet Just bought & sent ~0.7% of $MAX supply to the Hyperliquid Labs address for 25 $HYPE → hyperevmscan.io/tx/0xc9989f495… They now hold ~5.2% of the total supply Maxliquid.
Max tweet media
English
2
6
19
757
OpenServ
OpenServ@openservai·
Greg Ivanov, ex-Google, joined SERV as a key advisor. We are officially announcing his term on Spaces today at 1 PM EST. Join to find out where SERV Reasoning is heading and how it scales. Set your reminders, you don't want to miss this 👇
OpenServ tweet media
English
12
34
204
7.3K