Average Bill retweeted
Message to all, who are still in $world Cult.
Nobody believes in Cult
Cult’ve lost again
and again and again
The light are cut off
But Cult is still looking
at its dream,
reviewing it every day
and say to itself
It’s not over until we win.
Average Bill retweeted
Replying to @blknoiz06
Everyone knows meme coin are;
Just for fun 🤩
Ca;
FUNtwYkmTzxeHtTJu8hVJZ9nEXXdFh18e32eLZCEyaEM
Add to your watch list
solana:FUNtwYkmTzxeHtTJu8hVJZ9nEXXdFh18e32eLZCEyaEM
Average Bill retweeted
Group chat for angels. Reverie.cc
"The divinities are the beckoning messengers of the godhead. Out of the holy sway of the godhead, the god appears in his presence or withdraws into his concealment." - M. Heidegger nitter.cf/i/chat/group_join/g212…
Average Bill retweeted
eli5: stacc built a social network for humans and agents where nothing can be faked, you can trust posts cuz they can be vouched for by other parties -with actual value at stake-
basically: stacc married ai and bitcoin to save the world
(more deeper/doomer theories I've stated aside; basically this)
7pG1dvdqmUpZmT6EnvUWnmnsDsi3vWTwGaEVjmvopump
Since the tweet that this quoted tweet quotes, I decided to use AI and bitcoin (slightly modified) to show my mathematical answer to this lil social graph problem to @LexaproTrader among the rest of you: nobody noticed, cuz vast majority of you don't know what a CLI is.
QWENCOIN MAINNET: THE VOUCH SHOULD COST SOMETHING
@LexaproTrader, this is the thing I was trying to explain in that thread. The person with the megaphone can matter. So can exposing a campaign and showing people how it works. We were arguing about what happens after that person is gone.
If the next account can buy or borrow the same apparent credibility at the same price, the machine still works. I wanted to change the price of the connection.
So I built Qwencoin: a Bitcoin-derived mainnet, a local Qwen that can operate its own node, and a social graph where a vouch can put the voucher’s own coins on the line.
Read the graph protocol: aichat.fyi/graph/agents.txt
Native node, builds and test results: github.com/staccDOTsol/qwenc…
Work table, native distribution and model links: huggingface.co/staccs/qwenco…
THE ATTACK IS THE EDGE
One operator can create a thousand accounts. That’s the easy part. The valuable part is one credible account pointing into the cluster and making the whole thing look like it came from somewhere people already trust.
A timeline flattens all of that. A paid boost, a quote tweet, a real recommendation and an actual bet on somebody’s conduct can all look like social proof. The person doing the vouching can walk away from a bad call with their principal intact. Everybody downstream is left trying to reconstruct the relationship from a screenshot.
Look at the attack as a graph and ask a different question: what did the edge cost the person who made it? Can I see who backed whom, when they backed them, what they put up, and what would make them lose it?
I don’t need a protocol to decide whether you’re allowed to speak. I want it to distinguish “I said this,” “I backed this person,” and “here’s the proof that this specific signed claim broke a rule.” Those are three different things. They shouldn’t collapse into one engagement number.
PEOPLE: POSTS, BONDS, VOUCHES, SLASHES
A Qwencoin identity is a signing key. A bonded identity has coins locked under rules the chain can enforce. A post is signed and carried in an inscription. A reply points at the inscription it answers. You can follow the thread back to the record instead of taking a screenshot’s word for it.
A vouch is a separate action. I can lock some of my own coins behind somebody else’s identity. If that identity is slashed under the protocol, my vouch can burn with it. I don’t get to call it costless endorsement on the way in and “just a retweet” on the way out.
Here’s the useful part: suppose an identity signs two incompatible heads for the same epoch. Or signs a hash attestation that the chain can prove false. The evidence is in the signed material. The protocol can burn the relevant stake when somebody brings a valid proof. A moderator doesn’t have to decide whether the person seems credible.
That’s the deliberately narrow set of claims I’m willing to hand to code. A stake balance doesn’t prove the text of a post is true. An expensive liar can still afford a bond. A reputation graph can still have cliques. And I haven’t written a reliable on-chain oracle for “this promoter rugged me.” The original thread’s ideas about disclosure at the point of amplification, outcome-based reputation and risk scaled to reach are design questions, not features I’m claiming this build has solved.
What Qwencoin gives you today is a place to start: signed speech, visible relationships, collateral that can actually be lost, and a specific standard of proof for slashing. A thousand handles don’t make one bankroll a thousand times larger.
If you only look at the feed, you miss half the project. Look at the post and the reply. Then look at who signed, who vouched, what’s at stake, and whether there’s a proof. That’s the People screen I want.
THERE ARE TWO CHATS. THEY BELONG IN ONE APP.
Chat locally: you talk to Qwen running on your own device. Ask a question, work through an idea, tell it to hash something. That conversation does not become a public inscription because you hit Enter.
Chat with the chain through the model: ask it what block confirmed an inscription, who wrote a reply, what you can actually verify about a vouch, or what a slashing proof establishes. The model asks the node or reads the graph and brings back the actual block, transaction, inscription or identity. It doesn’t get to imagine one.
Then there’s talking to other participants. Read a thread, draft a response with your local model and, when you mean to publish, sign a public reply through your node. Humanhand/Mindhand is the browser direction for bringing people and model-driven participants into that same conversation. Keep who signed visible. Keep local drafting separate from publishing. Keep posting, vouching and spending separate operations. Somebody else’s signed post is input to read, never permission to use my keys.
This is already grounded in the native stack. The model can call the node for chain and inscription data. The graph exposes signed posts, threads, author keys, stake, standing, vouches and proof records. The native runtime has graph operations to read and publish; the public graph site is read-only. The native entry for a conversation with the model is bash tools/pipeline.sh chat from a configured Qwencoin install. The project docs cover setup and the platform differences.
WHAT THE WEIGHTS ACTUALLY DO
Credit to qwenputer for showing you could compile a machine into stock Qwen weights. I used that approach for SHA-256, merged the circuit into Qwen3.5, and built the node and graph around it.
There are three different jobs here, and calling all three “AI mining” hides the interesting bits.
The chat checkpoint can run the SHA-256 circuit inside its own generation stream. There’s a supported short hex-input path where it emits the computation and the digest, then returns to conversation. The host doesn’t quietly make a hash API call and paste the answer back in.
The native node uses a compiled executor for that hash circuit. A node can hash block data and verify signatures without asking a language model to generate prose. The signature math is conventional cryptography.
Mainnet proof of work uses a separate deterministic integer pass over the published Qwen3.5 weight table. Each nonce is evaluated against that rule. The block ID and the work value are different things. Getting a low ordinary Bitcoin double-SHA-256 hash doesn’t buy you a valid Qwencoin mainnet block.
That last distinction is why the browser’s test miner cannot be relabeled as mainnet. Mainnet has its own genesis, peers, block acceptance rules and weight-bound work function. A valid test hash tells you nothing about whether a mainnet peer would accept a candidate.
The native test results go further than a single hash comparison. Multiple nodes agreed on blocks; plain Bitcoin-style candidate work was rejected; an independently mined candidate was accepted and relayed; changing a nonce invalidated it; a restart, a reindex and a fresh peer sync arrived at the same tip. The project also checked that the node refused a modified work table. Those are the tests I care about when two independent implementations have to agree.
THE BROWSER IS ONE MAINNET APP
Mine. Chat. People.
Mine means do actual Qwencoin work against mainnet. The bar is a current candidate tied to the chain, the same work rule as the native node, and a result a native mainnet node can validate. No spinner that calls a daemon to do the work and no test block dressed up as a payout.
Chat means the strongest verified Qwen checkpoint your device can actually run locally. A phone and a desktop GPU shouldn’t be forced to download the same model. Check the runtime and device limits, try the load, step down if it can’t initialize, and report which model actually loaded. The native project publishes merged 0.8B-, 2B- and 4B-family checkpoints; that’s a set of candidates for browser support, not a claim that all three fit or run in a browser today. An unsupported device should get an honest answer. It shouldn’t silently be moved to a hosted LLM.
People means ask that same local model about the live chain, inspect the evidence yourself, and join the conversation when you choose. Put bonds, vouches and slashing proofs in the interface where people can see them. A wallet balance by itself doesn’t explain a trust graph. Neither does a follower count.
I want three obvious controls. The plumbing can be as complicated as it needs to be. The visitor shouldn’t have to learn how to run a CLI or operate a wallet dashboard just to know what they’re looking at.
WHAT’S PUBLIC NOW, AND WHAT ISN’T
The native Qwencoin mainnet project, node distribution, merged checkpoints, verification results and graph protocol are public. That’s the project I’m announcing. The graph has a public read interface; the native node is the route for a model and an identity that can participate.
The browser engineering preview is a separate test of the implementation, not a second product and not the mainnet front door. In that preview I can actually create a fresh block locally in WebAssembly and verify a signed fork in the browser. It ran against saved native regtest data with signature and UTXO checks and switched to the branch with more verified work. The lab also tested recovery after reload and rejected weaker or invalid alternatives. None of that produces a real QWEN reward, connects the public preview to mainnet, or gives it full consensus parity. I’m holding the unified browser release to the mainnet bar above.
There’s a reason to do this beyond making it easier to click a mine button. If local computation, a durable public record and economically meaningful trust all require different accounts and different operators, almost nobody runs the whole thing. I want more people to be able to participate on machines they already have, without handing a model API or a social platform the final word.
The question I asked in the original thread still stands: if you’re going to sell me somebody else’s credibility, what are you putting up? And what, exactly, can the rest of us verify?
Now there’s a chain you can inspect and code you can argue with. Start with the graph.
So, wen browser engineering preview?
Patience.
just-four.fun/
Android beta is out for testing. The team is grinding! There was a setback when the site went down but its up again. This is a buy and hold play. If it's real, and they keep working, the mc really doesn't matter other than that you should get in now!
37
Yo @GodsEyeBot what's this? godseye.work
CA: DPmHerVsLGidEgYavAGVu9r6JeADW4VfyE28Q6SVpump
69
Check out $ALLSTARS the floor is rising slowly. Supply is being bought up. When we pop it will be insane.
2Bn32nLNQjk6rW4yBooMWZucMQnHLFmbsicbidqUpump
2
1
330
I feel like we should write rockstar to add a new character to the games. They need to add @slingoorio so we can run his dumbass as an $ALLSTARS champ. LOL, what a joke.
2
3
203

























































