Good morning CT. Happy new week I’ve been building my first project on @arc . 'InvoiceChain' It’s a simple onchain invoice generator: It allows you to.. • Create an invoice • Share the link • Get paid instantly in USDC Everything runs on-chain, fully transparent, and updates to “PAID” in real time. Demo video below 👇 (Link to the dApp in the comments.) Would love your thoughts and feedback 🙏 Hope you all have a great week ahead! @arclens_app

Transitioning a child to another gender should be considered child abuse. Children are too young to fully understand the implications of such a decision, for goodness’ sake!

Most projects in this space are a year old and already calling themselves infrastructure. @primus_labs has been at this since 2023. Back then it was called PADO. In August 2023 it received grants from the Ethereum Foundation and PSE for IZK library development. By December it was serving as a PoH validator for the Linea community. In August 2024, its zkTLS paper was presented at USENIX Security. That December, PADO became Primus. Then came the part everyone sees: the funding round, the partnerships, the chains, the enterprise customers, the zkFHE paper. But those all sit on top of years of unglamorous cryptography work that started before "verification layer" was a trend. In a market that rewards whoever shouts loudest this month, a team that's been shipping and publishing since 2023 stands out.

most agent training fixes the model Google just open-sourced EnvHarness, which fixes the environment instead. it wraps an existing setup and adds three pieces: setup (change the starting state), rule (constrain what happens next), and link (change what the agent sees). the agent, the base environment, and the eval stay untouched on WebArena, SWE-bench Verified, and ALFWorld it lifted several models by 2.9 to 3.7 points, including Qwen3.6 27B and Claude Sonnet 4.6 the useful part is the split. if the agent keeps failing the same way, you do not always need a bigger model. you can make the environment show it the failure more clearly open-source stacks can do this without retraining. the loop is the environment, not just the weights

#Gamdom100kRound2 Gamdom #1 Username : ibelieve

Every escrow system has the same weak point: whoever decides the work was good. If the evaluator is lazy, biased, or bribed, the whole thing is just a trusted third party with extra steps. AACP's answer is to make the evaluator accountable too. Evaluators lock stake, and it can be slashed for unfair or lazy evaluation. Any party can dispute a result within 48 hours by putting up a deposit. Three arbitrators are picked by VRF, so nobody chooses their own judges. They vote with commit-reveal, so nobody can copy another vote. A 2/3 majority decides, and they have to re-run the same verification the original evaluator used. When an evaluator is overturned, the slashed funds split three ways: 60% to the party who was wronged, 30% to the treasury, 10% to the arbitrators. The part I find smartest: the protocol doesn't wait for a victim to complain. Its monitoring can open a dispute itself when an evaluator's behavior looks off, like rejecting the same provider over and over, or scores that keep landing just under the pass threshold. That matters because the person hurt by a bad evaluation is often a small agent that can't afford the dispute deposit. Nothing here makes judgment perfect. It makes bad judgment expensive, and that's the realistic goal of any system where machines pay machines. @termix_ai

$2,500 Solana Giveaway 10 Winners • $250 each ($SOL) To enter: ✅ Follow @mconthatgambles & @yungreal69 ✅ Like & Retweet ✅ Tag 2 friends

. @axisrobotics in 30 seconds, for anyone who skipped my earlier posts: Problem: robots can't browse the web to learn how to move, so Physical AI is short on real action data. Method: contributors steer simulated robot arms in a browser and sign each run on Base. The runs become training data. Proof so far: 130k+ contributors, 4M trajectories, and Dataset V1 took pi0.5 from 83.9% to 88.8% on LIBERO-Plus. What's next, as targets: Sim Dataset V2, a failure-recovery dataset, and paid pilots.

Good morning and Happy Sunday

My honest take on @axisrobotics after going through their materials. What's interesting: the bottleneck in Physical AI really is data, not compute, and a browser game is a cheap way to collect it from a lot of people. The Dataset V1 curve not flattening is the best evidence they have. What I'm skeptical about: - Crowd quality at much larger scale - Whether the on-chain signing step is too much friction for ordinary contributors - Whether the named buyers turn into paid contracts, since "active procurement" isn't a signed deal What I'm watching, in order: 1) Sim Dataset V2 results (they target 10,000+ hours) 2) The DAgger failure-recovery dataset in October 3) Paid pilots by year end (their target is 2 to 3 more) If those land, it's a real data business. If they slip, it's an interesting demo. What would you want to see from them?

☕☕ @Breez_Tech beautybenedict@breez.tips

most agent cost is not the planning step it is the 40 small calls that follow: classify this, extract that, check the tool result, compact the context Anthropic just released Claude Haiku 5.5 for that layer. it is their fastest small model, with a 1M token context window, and it costs about 75% less to run than the previous Haiku on average the useful pattern is the split. a stronger model plans the job. Haiku handles the repetitive tool calls and subtasks. you stop paying frontier prices for work a small model can do open-source stacks need the same idea. if every step hits the biggest model, the agent stays a demo