GM {{ First Name}}

A couple of things I’ve been thinking about:

  • AI agents are poised for a comeback soon, this time with a much better product foundation.

  • If you're building with crypto-native users as your primary market, it will be tough to find sustained demand. The most interesting startups have products already geared towards infinite markets.

  • Robotics is advancing more rapidly than most people realise. I spoke with the OpenMind team this week, they’re building the “Android for robots”. More consolidated thoughts on robots soon

Meanwhile, Andrej Karpathy just sketched the cleanest roadmap for software’s next phase: Software 3.0

Stage one: handwritten code. Stage two: neural nets. Stage three: large language models you “program” by typing plain English.

Karpathy describes LLMs as a new operating system for the internet, still crude and closer to a 1960s mainframe than a smartphone. They run in the cloud, burn cash, and can break on bad prompts. These models shine and stumble in equal measure.

His bet: every serious app turns partly autonomous. The model tackles grunt work while the user reviews and tweaks, creating a rapid feedback loop.

His advice: learn classic coding, neural nets, and how to prompt LLMs well. Build good interfaces. Keep the AI under control.

It’s in line with a lot of what we’re hearing too: the silicon valley tech giants of the last decade are all extremely scared of being disrupted by AI agents.

🚨 But wait—

  1. Are you in our telegram channel? Cyra, our AI researcher, drops daily AI alpha.

  2. Catch our podcast summaries too, complete with detailed show notes.

  3. And if you want to jam with us directly, hop into our Discord

🔗 Matchain has officially launched its native token $MAT, now trading on major exchanges like Binance Alpha, Kraken, and KuCoin.

🎉 Warden Protocol and Kaito AI have launched a three-season event distributing over $1M and 2.5% of $WARD supply to top-ranked community members.

🚀 Redbrick ($BRIC) will soon be listed on Binance Alpha with trading and an airdrop opening June 21. Eligible users can claim the airdrop in two phases using Alpha Points.

AI mindshare peaked at 60% last September to January. It dropped to around 30% but is back at 36% now. Fresh models and agents from OpenAI, Claude, and Gemini could push interest even higher.

Souce: kaito.ai

The Virtuals ecosystem and its Genesis launchpad powered a large part of that recovery. We’d be keeping an eye out for something exciting coming up from Virtuals in the next 1-2 weeks. 👀

🔥 Matchain Daily Transactions Grow

Matchain recently hit 5.1M daily transactions and over 484M total, up 146% this year.

Matchain lives on BNB Chain’s Optimism stack and zeroes in on three jobs: bullet-proof AI, dirt-cheap scaling, and user-owned identity plus data.

The headline act is MatchID, a universal digital passport. Plug in your email or Telegram, bind it to your wallet, and stay private thanks to zero-knowledge proofs. You decide when to share data, and you get paid when you do.

Matchain also uses a staking method to prove users are real people, which keeps bots out and makes advertising more effective.

Brands and apps can tap anonymized insights for sharper personalization and ad targeting, without prying into raw user data.

Paris Saint-Germain is already in. The top-division French football club is using Matchain’s PlayerID so athletes own their performance stats and fans unlock token-gated rewards.

Source: en.psg.fr

🧑‍💻Yupp Raises $33M Led By A16Z

Yupp just broke stealth with a $33 million seed round led by a16z, with backing from some of the biggest names in AI and crypto.

So what’s the play? Turn human feedback into AI’s trust layer, and pay users to help make it smarter.

You enter a prompt, see responses from over 500 AI models, pick the best one, and give quick feedback. Every vote earns you credits, with users raking in up to $50 a month just for choosing which answers vibe better.

Votes aren’t just for show. Each one is signed, tracked on-chain, and feeds into Yupp’s public leaderboard, the VIBE Score. No hidden grading, no mystery benchmarks. Everyone can see which models are actually useful and which ones are just… confidently wrong.

Source: yupp.ai

It all runs on crypto rails, with instant payouts across Base and Solana. Transparent, borderless, and designed to resist manipulation.

The founders know the game. They’ve shipped consumer-scale ML at Twitter, Google, and Coinbase. They’re betting that open, crowdsourced AI evaluation beats closed labs and stale benchmarks.

In short: more choice, more trust, real rewards, and better AI for everyone.

Fundraises

  • Gradient snagged $10M from Pantera and Multicoin to kickstart a fully decentralized AI runtime, led by its Lattica and Parallax protocols for open, global infrastructure.

  • PublicAI also banked $10M, with stc Group and NEAR Foundation backing its play to pay people for real-time data checks and output steering, the firm’s “Human Layer” concept.

  • PrismaX, launched with $11M from a16z CSX to build decentralized data rails for autonomous robots, blending crowdsourced data, teleoperation, and on-device AI.

  • Sahara AI’s Buidlpad community sale drew 103,000 subscriptions across 118 countries and raised $8.5M (with 777% oversubscription). Fun fact: it was fueled largely by USD1 from World Liberty Financial.

Startup Updates

  • OpenGraph opened private access to its decentralized data-tagging playground. Users can now label images, draw boxes, and verify results on testnet.

  • Gensyn AI has open-sourced CheckFree, a fault-tolerant method for decentralized training that eliminates checkpoints and redundant compute, achieving up to 1.6x faster training with no loss in convergence.

  • Hyli’s public testnet is live, inviting users to try proof-powered apps and stress-test the network. Phase 2 will let developers deploy their own apps.

  • Tickr, one of the first native Nillion apps, went live as a marketplace for provable trading alpha. Traders sell their edge, followers can copy trades using private performance data.

  • GenLayer rolled out Asimov, an AI-driven testnet for its Intelligent Blockchain aimed at creating a decentralized legal system for machines.

  • Bagel Labs unveiled The First Batch, an early access program inviting developers and creators to shape its platform, access new features, earn credits, and join exclusive events.

  • DSV Fund has invested $300K in Macrocosmos to support three Bittensor subnets—Apex, Data Universe, and Mainframe—advancing live AI testing, real-time data, and decentralized protein folding.

  • Masa launched Open Data Store, offering free AI-ready data, one-click scrapers, and live insights for X. LinkedIn and TikTok integrations arrive soon.

  • IO rolled out Launch IO, a hackathon inviting developers to build autonomous agents with its Intelligence APIs. Builders can earn prizes and IO credits for shipping real bots and copilots.

AI Agents

  • Stratos is now live on Virtuals Protocol, automating on-chain options trading with AI agents. $STRATOS holders get performance fees, governance rights, and personalized trading agents.

  • TAO Agent, a Bittensor co-pilot, converts on-chain data, GitHub commits, and sentiment from more than 120 subnets into clear intel. Built on ReadyAI’s Subnet 33, the bot now lives on X.

  • Gaia and Mindscraft released Gaia Intern, a customizable AI assistant that fields community questions, replies on socials, writes content, and posts dev updates for the Gaia ecosystem.

Web2 AI

  • Google announced the general availability of Gemini 2.5 Pro and Flash models, plus a new 2.5 Flash Lite preview, offering improved cost-speed efficiency and a full 1M context window.

  • Midjourney dropped its first video model at $10 a month, turning any still into motion with a single prompt and inching toward live, interactive AI worlds.

  • Proactor debuted an AI aide that remembers chat history, scans real-time context, and acts before you’re done typing, serving students, workers, and engineers alike.

  • MIT unveiled SEAL, a framework that lets LLMs tweak their own weights with reinforcement learning and generate fresh study material on the fly.

Ever wondered how machines talk?

In part two of State of the Swarm, @ChappieonChain dives into how AI agents pick partners, share secrets, and build trust on their own.👇

State of the Swarm (II): The Language of Machines

How AI Agents Are Learning to Talk

chainofthought.xyz/p/state-of-the-swarm-ii-the-language-of-machines

🔥 Our Weekly Top 5

Zuck is trying to steal OpenAI staff

Do data moats actually matter?

AI relationships are getting way too real

Render Network powers a Netflix short film

The more I use chatGPT, the dumber I get?

That’s a wrap for this week! Got thoughts, feedback, or something cool to share? Just hit reply, we read it all.

Cheers,

Teng Yan & Issy

Login or Subscribe to participate

This newsletter is intended solely for educational purposes and does not constitute financial advice. It is not an endorsement to buy or sell assets or make financial decisions. Always conduct your own research and exercise caution when making investment choices.

Reply

or to participate

Keep Reading

No posts found