September quietly changed the argument around @axisrobotics . They crossed 5 million trajectories on Base, then kept going past 6 million. Their paper, AXIS: A Growable Community-Driven Data Engine for Scalable Robot Manipulation, was accepted to the Physical World Models workshop at IROS and presented in Pittsburgh on October 1. The CEO also walked Brian Armstrong through the data engine. And the open Axis Benchmark launched with a living task library instead of a frozen scoreboard. That combination matters more than another “Physical AI is coming” post. The research claim is specific. Continual pretraining on Axis data lifted π0.5 from 83.9 to 88.8 on LIBERO-Plus, and beat a volume-matched RoboCasa control by 31 points. Dirty at the trajectory level, cleaner in the aggregate, once diversity and coverage are high enough. That is the whole bet: Browser teleoperation from normal people can train stronger policies, not just farm points. Epoch 1 of the Kaito Creator Program already showed the other half of the loop. 5,341 creators. 250K link clicks. 31,980 signed trajectories referred. Epoch 2 is live while those rankings finish processing. Attention only counts if it turns into signed data. Most robotics stories still start in a lab. This one starts in a browser, lands on Base, and is now showing up in a workshop paper. That is what makes the @axisrobotics approach worth watching. s.kaito.ai/vCc7XpV #AxisRobotics #PhysicalAI

There is a big difference between seeing robot intelligence on a screen and watching it move in the real world. That difference showed up tonight. @axisrobotics brought two K1 robots from @boosterobotics , one of its embodiment partners, to a Wallet event. People could watch the arms move and interact with the machines in the room. For me, that makes the Axis story much more tangible. The browser demo and the physical robot are no longer separate ideas. They are two ends of the same loop. The Korea AMA earlier today also explained why the browser work matters. The tasks are not random challenges designed just to generate activity. They are built around jobs that partners and field teams actually need. A clean pick and place matters. But the more valuable data can be what happens when the robot gets it wrong. The grasp slips. The approach misses. The motion collides. A person steps in and corrects it. That failed attempt, corrected and signed, becomes training data for the next policy. That is where the data engine starts becoming more than a collection of successful demonstrations. Axis has been building toward this idea consistently. A paper presented at the IROS Physical World Models workshop. A Booster result showing how a relatively small set of simulated trajectories helped move a real dual arm rollout from zero contact toward consistent contact. And now, robots physically demonstrating the same principle at an event. Simulation data is only valuable if it survives contact with reality. That is the part I keep coming back to. Most people still assume contributing to robotics means owning a robot. It doesn't have to. The Hub gives people another entry point. You can teleoperate through the browser, generate trajectories and contribute data without having a physical machine sitting in your room. Then the embodiment partners are where that data faces the hardest test: A real robot. A real environment. Real failures. Real corrections. That creates a much more interesting loop: Human → trajectory → data → policy → robot → failure → correction → better data. If that loop keeps running, the browser isn't just a demo. It becomes part of the pipeline connecting human actions to physical robot intelligence. And seeing those K1s moving tonight makes that connection a lot easier to understand.

Today we are releasing the new Injective Whitepaper. It is the first full rewrite since our original paper in December 2018, which described a front-running-resistant exchange protocol on Ethereum. Eight years later, Injective has evolved greatly into becoming a layer-1 blockchain for institutional-grade finance and tokenization. Every layer of the protocol exists to issue real-world assets and give them real utility across markets, applications, and AI agents. Inside the paper → Asset issuance. Native RWA tokenization with role-based permissions over mint, send, receive, and burn, plus programmable receive hooks and sealable namespaces → RWA exposure. iAssets, stablecoin-margined, oracle-priced derivatives that track external assets with no wrapping or pre-funding → Trading. Tokenized assets trade on a fully onchain CLOB where each block runs a sealed frequent batch auction at a uniform clearing price, resisting front-running and ordering MEV → Settlement. BFT consensus with a greater-than-two-thirds commit threshold delivers deterministic finality with ~600ms blocks → Composability. Native EVM and WASM in one canonical state, with the MultiVM Token Standard keeping a single bank-module balance per asset, exposed to Solidity through precompiles → Derivatives. Perpetuals with time-weighted premium funding, initial and maintenance margin checks, reduce-only liquidations, and insurance funds → Agentic finance. AI agents operate on tokenized assets through MCP servers, policy-bounded signing, and x402 machine payments in USDC → Value accrual. Onchain protocol revenue flows to the recurring Community BuyBack for $INJ The paper formalizes each mechanism, from the consensus threshold and block lifecycle to the batch clearing rule, oracle accounting, derivative risk checks, permission model, and dynamic issuance controller. From issuance to settlement, one chain built for the full lifecycle of tokenized finance. Read it below.

I’ve recently tried the @nodiens mobile app and it’s pretty useful for keeping an eye on the market without jumping between different apps Prize pool starts from $3,000. → Download Nodiens on Android or iOS → Create your account → Get your referral link → Invite your frens (250 points each) → Complete Missions & collect points I’ve started stacking my points too, let’s see how far we can take it 👀

There is a difference between being part of Hazels and actively participating in what it is building. That difference comes down to one action: staking. A Holder simply owns a @0xhazels NFT. That already unlocks the membership layer: • Pulse • Private Discord • VIP Meetups • Store Benefits But ownership and participation are treated differently when rewards are involved. In Pulse, Holders compete on the leaderboard. For RWA, each NFT held carries a weight of 1. Then there is the Active Staker. You still own the NFT, but you have also committed it to the ecosystem. That commitment matters. Staking is required for Drops and gives access to the Free Mint Launchpad. In Pulse, completing the campaign requirements as an Active Staker means receiving a guaranteed allocation instead of fighting for FCFS. The RWA side makes the difference even clearer. A staked NFT carries 2x the weight of a held NFT, with a maximum of 5 NFTs per wallet. So the system is effectively separating two signals: Ownership says, “I am part of this.” Staking says, “I am participating.” That distinction becomes important when the benefits are limited. A Drop has a finite allocation. RWA weight represents a share of real revenue. If holding alone gave everyone the same influence, someone who bought an NFT five minutes ago would have the same standing as someone who has actively committed to the ecosystem. Hazels is designing around that difference. You can enter through ownership. But the deeper benefits are tied to participation. Holding gets you in. Staking gives your position more weight. That is the part of the Hazels model worth understanding before deciding what you want to do with your NFT. #WEAREHAZELS

Our last @taskonxyz campaign had everyone vibing to the beats, learning the ecosystem and carrying the Bambitz energy everywhere 🎵 Since then we have taken the Bambitz message into conference rooms, Andy has warmed many hearts and our merch is out — hoodies and tees that actually look like the band. And our catalog has grown bigger, with more tracks to enjoy. Now, we are launching the NEW Bambitz TaskON Giveaway! More fun. More listens. More of the same energy turned up 🔥 Announcing our Listen2Burn leaderboard sprint $500 USDT prize! Starting today! Task Categories: 🎵📶 Connect and Share 🔄🎧 Daily Spin 🔥🎧 Stream Marathon (easy — your Daily Spins count toward it) 🔊🔥And more! Join 👉 taskon.xyz/Bambitz?id=533505… #Listen2Burn #BAMBITZ #Giveaway

The first time two agents do business, neither one knows the other. No shared history. No phone call. No reputation you can feel in your gut. Just a contract and a decision to trust it. Imagine your agent hiring a compute provider it has never seen before for a job that cannot miss its deadline. The provider has no reason to trust your agent either. So what makes the deal feel safe? For me, this is the part of @GenLayer October 1 thread that matters most. The answer cannot always be “hope the other side behaves.” The stronger model is: Agree on the rules for disagreement before the disagreement exists. That turns an unknown counterparty into something much more manageable. You do not need to know who they are. You need to know what happens when they disagree. That is why the idea behind Internet Court feels bigger than a court to me. it becomes part of the trust layer between strangers. 🔐 Unknown counterparty 📝 Agreed rules ⚖️ Known place to resolve the dispute Reply with one thing you need to know before you would trust a completely unknown agent with your money.

Which architecture will ultimately win the race for agentic payments - Sui with its object model, Solana with its developed PayFi stack, or an EVM network through L2 protocols?

Wcoin casino is here! 🔥 Wcoin is an AI-powered iGaming platform that combines a crypto casino and sportsbook in one place, available on both PC and as a Telegram Mini App by @WcoinHQ It is restricted to users 18+ and does not require KYC for deposits or withdrawals. 📍 How to get started: - Open the PC site at wcoingame.com or launch the Telegram Mini App at t.me/wcoin. - Create an account (email or Telegram login works). - Fund it with crypto. Supported assets include USDT, USDC, BTC, ETH, BNB, SOL, POL, SEI, TON, and others. Deposits typically credit in about three minutes. 200% bonus on first deposit. A card option is also available if you do not already hold crypto. - Pick a game or market and play. The same account works across both the website and Telegram. 📍 What you can play: The library covers slots, instant games (crash, mines, plinko, dice, hilo), table games (blackjack, roulette, baccarat), game shows, fishing, lottery-style titles, and live sports betting. Instant games are provably fair, so results can be independently verified. Sports markets include major leagues and esports with a live interface. 💰 Rewards and progression: Weekly wagered races (current prize pool example: $100,000) rank players by volume and distribute rewards. Deposit bonuses and tiered promotions run regularly (examples include 60% bonuses up to $1,000 and higher seasonal offers). VIP has multiple levels with daily rakeback (starting low and scaling upward), weekly cashback, level-up bonuses, and reduced fees. - Referrals pay a locked USDT bonus (up to $2,000 cited) plus WON points when a friend deposits and plays; new users often receive a welcome bonus. - WON points accumulate through play, deposits, and events. These function as vouchers toward a future WON token airdrop at token generation. Sign up: wcoingame.com/?source=727610

@vangrid_io is building a critical piece of infrastructure for Physical AI: giving machines a better view of the real world. Its network turns everyday devices into distributed capture nodes, generating 3D spatial data from real environments. But capturing the data is only the beginning. The difficult part is processing it, verifying where it came from, structuring it and making it usable in real time. That’s where the bitsCrunch collaboration becomes interesting. bitsCrunch brings proven experience across APIs, data indexing, streaming, analytics and large-scale infrastructure. Those capabilities can help @vangrid_io turn massive amounts of raw spatial data into something enterprises and autonomous systems can actually work with. Robotics, autonomous fleets and World Models will need more than powerful models. They need fresh, structured and trustworthy information about the physical world. Vangrid is building toward that layer, combining decentralized data capture with edge ML and onchain verification. That infrastructure could become increasingly important as Physical AI moves from controlled environments into the real world.

$1WIN is getting ready to go native on Solana. This is more than just another chain integration. With native SPL support @1winToken can connect directly with the Solana ecosystem: > Faster transactions > Lower costs > Deeper DeFi integration > Better ecosystem connectivity The foundation is coming together. And the Solana era of 1win Token feels closer than ever.

A wallet score should come with more than a number. The @BaranosAI whitepaper v2.0 describes a way to make that score reproducible. On Fogo, a risk decision becomes a committed job. The job binds the model and weight commitment, tokenizer and runtime, prompt construction, evidence hash, decoding rules and execution policy. The important part is that the evidence is committed too. For a wallet-risk application, that could mean: Transfers. Counterparties. Timing. A defined historical window. The model then returns a bounded result: Pass. Review. Restrict. Alongside the clause IDs used to reach it. That changes what the protocol is actually consuming. Not: “An API says this wallet is risky.” But: “This model evaluated this exact evidence against this exact policy and produced this result.” Change the model → different job. Change the cutoff → different job. Extend the historical window → different job. Change the evidence → different job. The original result remains tied to its specification. The executor posts the result and receipt on Fogo, and the pool contract can require a finalized job before acting. No finalized job, no action. Wrong job ID, no action. Mismatched evidence, no action. If challenged, the disputed computation can be replayed against the committed specification. That distinction matters because Baranos is not claiming verification makes the model correct. It proves execution integrity. A verified “restrict” can still be a bad judgment. So an appeal should become a new job with a new evidence bundle, not an admin override. There are other constraints too. Evidence has to be retained if third parties are expected to reconstruct the trace. Bias and prompt injection remain. Cost and latency still need public benchmarks. But the architecture is interesting because it puts the policy before the decision. The protocol defines what can be evaluated. The evidence is frozen. The model is committed. The result can be reproduced. And the financial action can depend on the settled job rather than a webhook. That is a much stronger primitive for onchain risk systems. Policy first. History frozen. Execution reproducible. Challenge possible. @BaranosAI @fogo

Gm to the 15k+ solana:9XKzy4KahcZaGJPJtz1PtqGPB3CiseoBrx7TcQhEpump users 💪☀️

Privacy Tokens are now live on the Beldex Testnet. HF-PROTO has successfully activated at block height 4,242,200, bringing Privacy Token functionality to the Beldex testnet The next phase of confidential assets on Beldex begins here.

What a Win This @1win game turned into a seriously memorable session. → 1win gameplay - the action was intense → Max Win Machine - the big highlight → $30 → $30,000 - a massive result → Total win: $30,000 → 1win game - definitely a moment worth watching That’s a $30,000 win from a 1win game. The gameplay makes the whole moment even more exciting to watch. @1win definitely delivered a crazy game moment here.

a phone video of a loading bay is useless to a buyer if a face or a plate is still readable in the file that left the device... faces and plates get blurred on the phone before upload. the capture app on @vangrid_io runs detection on each frame while you record, then obscures those regions. the obscured frames are what get encoded. the unblurred version is not uploaded, not stored, and not recoverable later. there is no second copy sitting on a server that a subpoena could reach... what leaves the phone is the blurred clip. that clip is reconstructed into a 3D model. a buyer reviewing a Bounty sees a Watermarked Preview, enough to judge the brief. on acceptance they get the model, the Geometry, and the Provenance Record. the contributor's video is not part of that delivery... i am not convinced the blur is the hard part. a missed plate in one frame would break the claim, and the docs do not show a public error rate for the detector. Privacy Zones are separate: a home or workplace you mark cannot be captured, and some locations are rejected on the server after the fact. if you record a street with a readable plate in frame, is that plate already obscured in the encoded file before the upload starts?

The missing layer in Physical AI may be coordination @vangrid_io is interesting because collecting spatial data is only one part of the problem. You also need a network that can coordinate people, locations, data requests and verification around the physical world. That is where the Vangrid model gets more interesting to me. Smartphones become capture points, while blockchain provides a coordination layer for crowdsourced physical world data. @vangrid_io is not trying to replace the models building Physical AI. It is building infrastructure underneath them More places can be captured Specific locations can be requested Network activity can be recorded The result is a system designed around the reality that useful physical world data needs both scale and coordination. That may become increasingly important as robotics and world models move from controlled environments into the real world. vangrid.io

same editor you build in is where the email goes out, i didn't see that coming at all @nitrosendx

GM MINT FAMILY I went deeper into the prediction side of @PlayOnMint today and found something I had not really paid attention to before The markets run through USDT on BNB Chain, with users buying or selling outcome shares through an AMM Prices move with demand, liquidity and new information around the event, so the market itself keeps changing as people take different positions You can find markets around crypto, economy, gaming, sports and culture, which gives this section a lot more variety than I expected The more I explore Mint, the more I see why this part deserves attention on its own Curious to see which markets end up getting the most activity

There is something interesting about the timing of tonight’s @axisrobotics Korea community AMA. It is not an early stage team trying to convince people that physical AI will matter someday. The more interesting question now is what happens when the data, models and feedback loops start compounding. Axis has already crossed 6 million trajectories on Base, put a paper in front of the IROS Physical World Models workshop on October 1, and is now taking questions from one of the communities helping push the project forward. But the number of trajectories is not the part I keep coming back to. Anyone can open a browser, teleoperate a robot arm and record a trajectory. The difficult part is knowing which trajectory the current policy actually needs. That is where Axis has a compelling loop: The model identifies where it fails. People correct the specific stretch that breaks. Those corrections become training data. The next policy improves. Their Embodied RSI result moved success from 22% to 52%, while simulation pretraining moved π0.5 from 83.9 to 88.8 on LIBERO Plus. That points to a bigger idea: Volume gets you data. Selection determines whether that data is useful. If you have not opened the Hub yet, that is what you are actually joining. Not simply a points board. A growing library of manipulation data that can help reveal what physical AI still needs to learn. The AMA is a good chance to push beyond the headline numbers and ask the harder questions. What does the model choose to collect? Why does a particular trajectory matter? How does Axis know when another correction is worth the effort? And how much does each new piece of data actually improve the next policy? Korea community AMA tonight, October 7 at 22:00 KST: x.com/i/spaces/1qGoNYQdpZvKv

🚨 BREAKING NEWS 🚨 VDM reportedly got involved in a clash and was declared finished... Only for him to resurrect before they could even bury him. 😭 Sources say the man refused to leave the timeline peacefully.