What happens when you post a real Monet and say it’s AI? The coolest art social experiment I’ve seen in a while. Thank you @SHL0MS

rsETH protocol operations have fully resumed. - The exchange rate was updated at 16:45 CET today, reflecting all staking rewards accrued during the pause. - EIGEN claims are now distributable to rsETH holders for the full pause window. - Deposits and withdrawals are live on mainnet and L2’s - rsETH backing is fully restored. Details below. 🧵

If your favourite CT says there’s nothing interesting to talk about because the market is dead, hit that unfollow button. They’re washed and can’t do proper research. There are a ton of hidden gems in this space, for example, @prlnet, which is flying under everyone’s radar. Here’s my thesis breakdown in under 30s: - ➠ What Pearl Is Pearl is a Layer 1 blockchain that forks Bitcoin and replaces SHA-256 with arbitrary matrix multiplication as its proof-of-work mechanism. In other words: it's a Bitcoin clone where miners run AI math instead of useless hashing puzzles. The same GPU that answers a ChatGPT query can also mint the token and secure the network at the same time. - ➠ The Problem It Solves Right now, AI runs on a winner-takes-all stack. You pay OpenAI, Anthropic, or a hyperscaler for compute. They own the datacenter. They keep the margin. The electricity gets burned, the model gets smarter, and the upside belongs to whoever owns the rack. The rest of the world has no way to own a piece of "AI compute" as a category. There's no neutral asset that tracks the cost of GPU work the way gold tracks the cost of pulling metal out of the ground. Meanwhile, Bitcoin has the opposite problem. Bitcoin's proof-of-work is brilliant security but the actual computation (SHA-256 hashing) produces nothing. Miners burn billions of dollars of electricity solving puzzles whose only purpose is to be hard to solve. Pearl's bet: what if the puzzle that secures the chain was also the puzzle the rest of the economy already wants solved? - ➠ How it Works Every modern AI workload (inference, training, fine-tuning, embeddings) boils down to one operation repeated billions of times: matrix multiplication (MatMul). Two grids of numbers multiplied together. That's what GPUs are built for. Pearl replaces Bitcoin's SHA-256 with arbitrary matrix multiplication as the proof-of-work. The math paper from Komargodski (Hebrew University) and Weinstein (Hebrew University / ex-Columbia, ex-NVIDIA, ex-VAST Data) proves you can make any MatMul cryptographically verifiable with overhead of 1 + o(1) - math notation for "negligible." Concretely: if running an inference normally costs you 100 GPU-cycles, running the same inference in a way that also produces a valid Pearl block costs you maybe 100.5 cycles. The security tax is rounding error. So the same kilowatt-hour does three jobs at once: ❶ Answers the customer's AI query (the useful work) ❷ Earns the miner a PRL block reward (the monetary incentive) ❸ Secures the Pearl blockchain (the network effect) This is what "proof-of-useful-work" (PoUW) means. Bitcoin's PoW is wasteful by design. Pearl's PoW is productive by design. - ➠ Why The Token Has A Price Floor The whitepaper boils down to the rule miners already follow: cost to mine = value of reward If PRL trades below production cost, miners shut off and difficulty falls. If it trades above, miners pile in and difficulty rises. Standard PoW. What’s different is the “cost.” Bitcoin’s is electricity. Pearl’s is useful GPU compute, the scarcest resource right now. So PRL’s price floor tracks the global AI compute cost curve. As long as H100/H200 supply is tight and capex keeps rising, producing PRL gets more expensive. - ➠ Tokenomics ▸ Total supply: 2.1B PRL forever ▸ Premine: nothing ▸ Team allocation: nothing ▸ Public sale: never happened ▸ Emission curve: smooth daily decline (not Bitcoin's halving cliffs) Bitcoin halvings hit miners hard every four years: revenue drops 50% overnight, inefficient miners wash out, and the network reshuffles. Pearl smooths this into a daily decline. Revenue fades gradually, making capacity planning far more predictable. Roughly half the supply is issued in the first four years, then issuance slows toward the cap. - ➠ The Current State ▸ Mainnet launched April 27 ▸ Hashrate 3.56 EH/s (~0.4% of Bitcoin) ▸ ~100s blocks (difficulty lagging miner growth) ▸ H100+ only: no gaming GPUs, no hobbyists ▸ OTC: $0.30 → $0.70 in 3 days ▸ Cost to produce 1 PRL: $0.30 → $0.51 ▸ Subsidized compute pool: $0.004 → $0.0052 per TH-hr (5 days) No CEX or DEX, no CoinGecko/CMC. token is live, but there’s basically no chart for retail to trade. - ➠ Future Lookout Right now people buy $PRL because mining costs are rising while supply is fixed. The real pivot is the planned compute marketplace (not live yet): - Pay for inference in PRL, not dollars (vs AWS) - Reserve/settle GPU contracts on-chain in PRL - Verify compute cryptographically, not via cloud billing If it ships and works, PRL shifts from “token tracking compute costs” to “unit of account for AI compute.” USDC did this for DeFi, one standard token replaced messy, fragmented settlement. PRL is aiming to be the same for compute. FYI, Render, Akash, and even Bittensor are pursuing similar ideas, but no one has won yet. - ➠ Institutional Locked-In ▸ a16z crypto hosted a talk on the Pearl paper. The page got pulled (now 404s) and archived. Pulled pages usually signal private conversations starting, not interest dying. ▸ Stanford Crypto Seminar hosted Weinstein. Same room that hosted the foundational talks for Bitcoin and Filecoin. ▸ Rob Hadick at Dragonfly (the GP who led Polymarket, Ethena, and Rain) publicly follows the account. ▸ Yonatan Sompolinsky (Kaspa founder, the most important PoW theorist outside Bitcoin Core) is in the paper's acknowledgements. That's not a polite footnote. That's the closest thing to a blessing the PoW research world produces. No VC has publicly invested. Given the no-premine / no-team-allocation structure, there's nothing for VCs to buy at a discount. They'd have to mine like everyone else. - ➠ Bear Scenario for PRL → The price jumped 133% in over-the-counter trading, but there weren’t many real trades. because trading was thin, we don’t know if the price would hold up when lots of buyers and sellers show up. → The “compute marketplace” is mostly a plan right now. if they don’t actually build it or it doesn’t work, then the project is basically just another bitcoin-like coin with a different math story. → They say there was “no premine” (no big stash made for insiders at launch), but no independent audit has confirmed it. for now you’re taking the team at their word until someone analyzes the distribution. → The low mining/pool rates are low because they’re being subsidized. once the subsidies end, the costs and incentives might change and the current economics might not hold. → The core crypto claim (verifying large matrix multiplications securely, even if someone is trying to cheat) is genuinely novel, but the efficiency claim is proven in theory. it hasn’t been proven in the real world under heavy load and active attacks. - ➠ Final Take Pearl makes AI math do double duty as money math. If that idea works at scale, the token becomes the most economically grounded crypto asset ever shipped, because its production cost is tied to the most demanded resource in the world. If it doesn't work, well, you get a well-designed Bitcoin clone that nobody uses.

We’re restructuring Dune to sharpen our focus around the core data products thousands of customers across the crypto industry rely on. That unfortunately means we’ve let 25% of the team go this week. These are exceptional people I can wholeheartedly recommend — ping me if you’re hiring top crypto talent. Going forward, Dune is all-in on two shifts: AI and institutions coming onchain. All the way back in 2018 Dune pioneered crypto data accessibility, and we’re excited to accelerate that faster than ever with AI. We’re the only player who has done the hard work of building the end-to-end stack for crypto data: ingestion, QA, storage, cleaning, normalizing and querying. With Dune MCP, teams and agents can now build dashboards and workflows without needing to know anything about SQL nor data infrastructure (and associated bills). Dune already serves virtually every leading crypto company, and now also the world’s most forward-leaning financial institutions. As currencies, stocks, bonds, commodities and more move onchain, we’re investing heavily in our data layer and white-glove service to power institutions. For 8 years we've grown through multiple rollercoaster cycles while other data providers have come and gone. We remain well capitalized, excited about the future, and committed to our mission of making crypto data accessible. The Data Must Flow.