Q2 @Mable_Jiang:As of today, many L2 solutions have emerged, primarily based on the OP stack, along with some attempts at zkrollups. I would love to hear your evaluation of the rollups path over the past few years, with the hope that it will be as objective as possible:What do you think has been done well, and what differs from your expectations back then?Are rollups overall good for Ethereum, or are they siphoning value (I saw you calling for these L2s to give back to Ethereum a few days ago)?Does ETH really need these L2s?https://t.co/9tw6ovHeKQA @VitalikButerin:So far, our scaling approach can be understood as hybrid L1 + L2, but I think no one has clearly defined which transactions should be on L1 and which should be on L2.The answer of "put everything on L2" is hard to accept because:* This risks losing ETH's roles as a medium of exchange, store of value, etc. If you worry that L2 is stealing users from L1 without providing anything back to L1, this issue would be even more serious in a situation where "L1 doesn't do much at all".* Cross-L2 operations still require L1. If an L2 has issues, users still need a way to move to another L2. Therefore, there are some use cases where L1 cannot be avoided. I have written an article on this topic here: https://t.co/Kr5rfcdY0KThe answer of "put everything on L1" is also hard to accept because:* If L1 supports many transactions, it can easily become centralized, even with technologies like ZK-EVM, etc.* The demand for on-chain transactions is infinite. No matter how high L1's TPS is, there will always be an application that needs 10 times more TPS (for example, artificial intelligence, micropayments, small prediction markets, etc.).* L2s do more than just scaling; they can also provide faster confirmation speeds through pre-confirmations and avoid MEV issues through sequencers.So we need hybrid L1 + L2. I think the role of L2 will continue to change, for instance, it seems that EVM-equivalent L2s are already sufficient; we might see more L2s focused on privacy (like Aztec, Intmax, etc.), and possibly more application-specific L2s (if an application wants to control its MEV situation, there are benefits, etc.).In the short term, I think we should continue to improve L1's capabilities, increase blobs to give L2 more space, promote interoperability across L2s, and then the market will decide which scaling method is suitable for which applications.https://t.co/vISY0yN7b1