yezhang1998
0 Suivis
0 Abonnés
2 J’aime
0 Partagé(s)
Publications
I was just chatting with a friend about AI vs crypto, and it reminded me of a personal story. I actually made my choice between the two seven years ago during undergrad.
I was doing my thesis in a computer architecture lab. There were two directions I could take:
1) Hardware acceleration for AI — a hot, well-established field with lots of prior work and support
2) Hardware acceleration for crypto — a niche area, barely anyone working on it, especially not zk
I started with AI. I read the ML books, played with models... and quickly got lost. I couldn’t explain anything. You train a model, tweak some random parameters (epochs, batch sizes, whatever), and out pop some weights. Sure, there’s math behind it, but it often feels more empirical than principled. Trial and error. Black magic. I don't know if OpenAI could have predicted how good GPT would become before it actually worked that well.
That lack of mathematical intuition bugged me. I’ve always needed to understand things deeply, to reason about systems, not just observe them. So I switched to the second path.
I went all-in on zk, from scratch. I was literally the only one working on it in my department. It was brutally hard, but it felt right. The math in cryptography is fundamentally different. You deal with polynomials, probabilistic proofs, number theory, all things that give you strong, provable guarantees. You know why something works. You can trace every step. There's beauty and certainty in that.
Maybe I’m just dumb when it comes to AI — I still don’t "get" how those weights emerge from training. And yes, things have changed with LLMs and AGI becoming real, and I’m definitely curious about the intersection. But regardless, to me, crypto still feels like pure science, elegant and beautiful :)
I was doing my thesis in a computer architecture lab. There were two directions I could take:
1) Hardware acceleration for AI — a hot, well-established field with lots of prior work and support
2) Hardware acceleration for crypto — a niche area, barely anyone working on it, especially not zk
I started with AI. I read the ML books, played with models... and quickly got lost. I couldn’t explain anything. You train a model, tweak some random parameters (epochs, batch sizes, whatever), and out pop some weights. Sure, there’s math behind it, but it often feels more empirical than principled. Trial and error. Black magic. I don't know if OpenAI could have predicted how good GPT would become before it actually worked that well.
That lack of mathematical intuition bugged me. I’ve always needed to understand things deeply, to reason about systems, not just observe them. So I switched to the second path.
I went all-in on zk, from scratch. I was literally the only one working on it in my department. It was brutally hard, but it felt right. The math in cryptography is fundamentally different. You deal with polynomials, probabilistic proofs, number theory, all things that give you strong, provable guarantees. You know why something works. You can trace every step. There's beauty and certainty in that.
Maybe I’m just dumb when it comes to AI — I still don’t "get" how those weights emerge from training. And yes, things have changed with LLMs and AGI becoming real, and I’m definitely curious about the intersection. But regardless, to me, crypto still feels like pure science, elegant and beautiful :)
It’s impressive that every hack eventually leads to buying ETH…
Even the hackers know which chain is truly decentralized and secure, they actually care, since they could lose everything or end up in jail.
Even the hackers know which chain is truly decentralized and secure, they actually care, since they could lose everything or end up in jail.
I’m increasingly convinced that Validium-style systems will be key to institutional adoption. It offers:
1) Regulation-friendly privacy at extremely low cost
2) Trustless, permissionless interoperability with other systems
ZK enables systems to "talk" to each other in a verifiable and privacy-preserving way.
Imagine untrusted banks or fintechs needing to clear and settle — the choice is either a closed, permissioned system (like Hyperledger), or a verifiable, privacy-preserving alternative. Even SWIFT is a system with closed membership.
Validiums are the only practical path to the latter. They're the only way to let systems interact securely without needing to trust each other. (Public chain destroys privacy or efficiency, or will have tons of issues with the regulatory side)
1) Regulation-friendly privacy at extremely low cost
2) Trustless, permissionless interoperability with other systems
ZK enables systems to "talk" to each other in a verifiable and privacy-preserving way.
Imagine untrusted banks or fintechs needing to clear and settle — the choice is either a closed, permissioned system (like Hyperledger), or a verifiable, privacy-preserving alternative. Even SWIFT is a system with closed membership.
Validiums are the only practical path to the latter. They're the only way to let systems interact securely without needing to trust each other. (Public chain destroys privacy or efficiency, or will have tons of issues with the regulatory side)
My take on why REV is flawed as a valuation metric for ETH:
1. REV is a vanity metric, like TVL
It’s useful to track, but the number alone doesn’t mean much. Take TVL as an example: 10M USDT from one whale is different from 1M users with $10 each, the latter is far more robust. Like TVL, REV doesn’t capture user diversity, retention, or app quality. It’s a shallow metric without context.
2. REV shouldn't be maximized long-term
REV is fundamentally a cost to users. As apps internalize MEV and fees approach zero (especially with app-specific sequencers), chains with high REV may simply be inefficient.
The most decentralized and credible global settlement layer should aim to secure the most valuable outcomes, not extract the most fees. It will be higher as a valuable economic zone, but REV is just one of many metrics.
3. ETH derives value from network confidence and trust (agree with @ryanberckmans)
Key pillars of ETH’s value:
> Deep liquidity and scalable social consensus
> The most secure base for rollups and modular apps
> ~$225B in app capital
> ~80% share of real-world assets
Other examples: Tron has more REV than Solana but 1/10 the market cap. BTC has almost no fees but is worth $2T. ETH may not be a pure SoV like BTC, but it’s a highly liquid, trusted, and programmable asset. Every rollup settling on Ethereum reinforces its value. Unless another L1 can replicate ETH’s economic weight and systemic trust, there’s no better foundation for an open crypto economy.
The ticker is ETH.
Other recommend read
1. From @ryanberckmans
https://t.co/7jRYuvGTDI
2. From @sassal0x
1. REV is a vanity metric, like TVL
It’s useful to track, but the number alone doesn’t mean much. Take TVL as an example: 10M USDT from one whale is different from 1M users with $10 each, the latter is far more robust. Like TVL, REV doesn’t capture user diversity, retention, or app quality. It’s a shallow metric without context.
2. REV shouldn't be maximized long-term
REV is fundamentally a cost to users. As apps internalize MEV and fees approach zero (especially with app-specific sequencers), chains with high REV may simply be inefficient.
The most decentralized and credible global settlement layer should aim to secure the most valuable outcomes, not extract the most fees. It will be higher as a valuable economic zone, but REV is just one of many metrics.
3. ETH derives value from network confidence and trust (agree with @ryanberckmans)
Key pillars of ETH’s value:
> Deep liquidity and scalable social consensus
> The most secure base for rollups and modular apps
> ~$225B in app capital
> ~80% share of real-world assets
Other examples: Tron has more REV than Solana but 1/10 the market cap. BTC has almost no fees but is worth $2T. ETH may not be a pure SoV like BTC, but it’s a highly liquid, trusted, and programmable asset. Every rollup settling on Ethereum reinforces its value. Unless another L1 can replicate ETH’s economic weight and systemic trust, there’s no better foundation for an open crypto economy.
The ticker is ETH.
Other recommend read
1. From @ryanberckmans
https://t.co/7jRYuvGTDI
2. From @sassal0x
(not so) hot take:
Traditional payment companies don’t actually want irreversible instant finality. What they need is fast, controllable pre-confirmation.
Legacy systems rely on layers like SWIFT, correspondent banks, and clearing houses to handle compliance, fraud checks, and settlement. The delays are intentional.
Crypto’s real value is removing many of those middlemen, but not all. To onboard institutions at scale, we need compliant, auditable, and faster infrastructure without forcing everything fully onchain from day one.
Traditional payment companies don’t actually want irreversible instant finality. What they need is fast, controllable pre-confirmation.
Legacy systems rely on layers like SWIFT, correspondent banks, and clearing houses to handle compliance, fraud checks, and settlement. The delays are intentional.
Crypto’s real value is removing many of those middlemen, but not all. To onboard institutions at scale, we need compliant, auditable, and faster infrastructure without forcing everything fully onchain from day one.
One super interesting take from a conversation with @drakefjustin:
Bitcoin’s security model will be broken in 10 years. ETH will be the only money left — ultrasound, credibly neutral, decentralized, and economically sound.
Here’s a deeper dive into why Bitcoin’s model might not hold up:
> Bitcoin relies on miners for security. Today, miners are paid through a block subsidy (newly minted BTC) and transaction fees. The subsidy is the dominant component, accounting for more than 90% of miner revenue.
> But every four years, the subsidy is cut in half. This means Bitcoin’s native inflation, which effectively funds its security, declines exponentially. Eventually, the network will rely entirely on transaction fees to incentivize miners. There is no guarantee that fees alone will be enough to maintain adequate security.
> Some argue that BTC's price will increase and make up for the lost subsidy. But if you run the numbers, the flaw becomes obvious. As the price of BTC rises, the value that miners are securing also rises, often by a much larger factor. So you're effectively paying less and less to secure more and more value.
This isn’t FUD. It’s a real, long-term economic question baked into the protocol’s design.
Hard to imagine what crypto will look like in 10 years. But one thing feels increasingly likely:
The ticker is ETH. 🦇🔊
Bitcoin’s security model will be broken in 10 years. ETH will be the only money left — ultrasound, credibly neutral, decentralized, and economically sound.
Here’s a deeper dive into why Bitcoin’s model might not hold up:
> Bitcoin relies on miners for security. Today, miners are paid through a block subsidy (newly minted BTC) and transaction fees. The subsidy is the dominant component, accounting for more than 90% of miner revenue.
> But every four years, the subsidy is cut in half. This means Bitcoin’s native inflation, which effectively funds its security, declines exponentially. Eventually, the network will rely entirely on transaction fees to incentivize miners. There is no guarantee that fees alone will be enough to maintain adequate security.
> Some argue that BTC's price will increase and make up for the lost subsidy. But if you run the numbers, the flaw becomes obvious. As the price of BTC rises, the value that miners are securing also rises, often by a much larger factor. So you're effectively paying less and less to secure more and more value.
This isn’t FUD. It’s a real, long-term economic question baked into the protocol’s design.
Hard to imagine what crypto will look like in 10 years. But one thing feels increasingly likely:
The ticker is ETH. 🦇🔊
One super interesting take from a conversation with @drakefjustin:
Bitcoin’s security model will be broken in 10 years. ETH will be the only money left — ultrasound, credibly neutral, decentralized, and economically sound.
Here’s a deeper dive into why Bitcoin’s model might not hold up:
> Bitcoin relies on miners for security. Today, miners are paid through a block subsidy (newly minted BTC) and transaction fees. The subsidy is the dominant component, accounting for more than 90% of miner revenue.
> But every four years, the subsidy is cut in half. This means Bitcoin’s native inflation, which effectively funds its security, declines exponentially. Eventually, the network will rely entirely on transaction fees to incentivize miners. There is no guarantee that fees alone will be enough to maintain adequate security.
> Some argue that BTC's price will increase and make up for the lost subsidy. But if you run the numbers, the flaw becomes obvious. As the price of BTC rises, the value that miners are securing also rises, often by a much larger factor. So you're effectively paying less and less to secure more and more value.
This isn’t FUD. It’s a real, long-term economic question baked into the protocol’s design.
Hard to imagine what crypto will look like in 10 years. But one thing feels increasingly likely:
The ticker is ETH. 🦇🔊
Bitcoin’s security model will be broken in 10 years. ETH will be the only money left — ultrasound, credibly neutral, decentralized, and economically sound.
Here’s a deeper dive into why Bitcoin’s model might not hold up:
> Bitcoin relies on miners for security. Today, miners are paid through a block subsidy (newly minted BTC) and transaction fees. The subsidy is the dominant component, accounting for more than 90% of miner revenue.
> But every four years, the subsidy is cut in half. This means Bitcoin’s native inflation, which effectively funds its security, declines exponentially. Eventually, the network will rely entirely on transaction fees to incentivize miners. There is no guarantee that fees alone will be enough to maintain adequate security.
> Some argue that BTC's price will increase and make up for the lost subsidy. But if you run the numbers, the flaw becomes obvious. As the price of BTC rises, the value that miners are securing also rises, often by a much larger factor. So you're effectively paying less and less to secure more and more value.
This isn’t FUD. It’s a real, long-term economic question baked into the protocol’s design.
Hard to imagine what crypto will look like in 10 years. But one thing feels increasingly likely:
The ticker is ETH. 🦇🔊
说起来刚开始搞scroll的时候也是22岁，是我刚毕业的第一份正经事业
真的是时光荏苒
不知不觉已经四年了
一路踩了不少坑，学到很多东西，未来的路也还长，要持续学习，持续build
哦，再之前本科还在conflux实习过，0.8抄底了cfx🥹能找主任维权么@forgivenever
真的是时光荏苒
不知不觉已经四年了
一路踩了不少坑，学到很多东西，未来的路也还长，要持续学习，持续build
哦，再之前本科还在conflux实习过，0.8抄底了cfx🥹能找主任维权么@forgivenever
Honestly, I still can’t tell if @Optimism and @0xPolygon are truly collaborating or just throwing the nicest shade ever...
Huge respect to both teams:
1. Props to the OP team for building and open-sourcing a modular rollup stack framework
2. Props to the Polygon team for advancing open-source zk proving infrastructure
That said, if the endgame is an interop layer, I do believe the zk-based approach has a better shot at winning on scalability. The Superchain model comes with too many trade-offs beyond the 15% tax (i.e., limited chain customization, requirement to run a follower node for each other). And frankly, both teams need to ship faster.
Some of my personal thoughts on interop:
1. Interop is overrated. Most issues can be addressed via intent-based bridges. Interop matters more to long-tail or app-specific rollups to access liquidity and users from larger, general-purpose rollups that have their own economic zones (e.g., Base).
2. Chains will increasingly need deep customization, it will make seamless interop hard. Even if you only look at sequencing, @unichain uses TEE-based sequencing, @Celo has its own consensus, some chains want to be based, others want real-time pre-confirmation. These divergences will always exist. You can try to mask it all behind “chain abstraction,” but that comes with heavy trade-offs.
3. I see the Superchain more as a social and relational network than a technical moat. Without a new kind of moat, it’ll be hard to coordinate and govern such a diverse chain ecosystem effectively in the very long term.
Huge respect to both teams:
1. Props to the OP team for building and open-sourcing a modular rollup stack framework
2. Props to the Polygon team for advancing open-source zk proving infrastructure
That said, if the endgame is an interop layer, I do believe the zk-based approach has a better shot at winning on scalability. The Superchain model comes with too many trade-offs beyond the 15% tax (i.e., limited chain customization, requirement to run a follower node for each other). And frankly, both teams need to ship faster.
Some of my personal thoughts on interop:
1. Interop is overrated. Most issues can be addressed via intent-based bridges. Interop matters more to long-tail or app-specific rollups to access liquidity and users from larger, general-purpose rollups that have their own economic zones (e.g., Base).
2. Chains will increasingly need deep customization, it will make seamless interop hard. Even if you only look at sequencing, @unichain uses TEE-based sequencing, @Celo has its own consensus, some chains want to be based, others want real-time pre-confirmation. These divergences will always exist. You can try to mask it all behind “chain abstraction,” but that comes with heavy trade-offs.
3. I see the Superchain more as a social and relational network than a technical moat. Without a new kind of moat, it’ll be hard to coordinate and govern such a diverse chain ecosystem effectively in the very long term.
The whole conversation is like
Fede: Ethereum leadership failed to work with Starkware and almost abandoned them. Need to involve Eli in decision-making.
Me: I don’t see Starkware being treated unfairly. Their tech stack is different, but they’re treated like everyone else. This is a broader issue, and we need to involve more legit, tech-driven Layer 2s in the process.
Fede: You just don’t understand how important Starkware is. Let me teach you with (wrong facts). Others can’t compare to Starkware.
Me: Dismissing others' contributions with incorrect facts isn’t the way forward. I’m not saying Starkware isn’t important, just that they’re not treated any differently, and leadership needs to be more inclusive and supportive of tech teams.
Fede: You’re flexing credentials, I was gay, I know more about butt hurt.
Me: ....👍
Fede: Ethereum leadership failed to work with Starkware and almost abandoned them. Need to involve Eli in decision-making.
Me: I don’t see Starkware being treated unfairly. Their tech stack is different, but they’re treated like everyone else. This is a broader issue, and we need to involve more legit, tech-driven Layer 2s in the process.
Fede: You just don’t understand how important Starkware is. Let me teach you with (wrong facts). Others can’t compare to Starkware.
Me: Dismissing others' contributions with incorrect facts isn’t the way forward. I’m not saying Starkware isn’t important, just that they’re not treated any differently, and leadership needs to be more inclusive and supportive of tech teams.
Fede: You’re flexing credentials, I was gay, I know more about butt hurt.
Me: ....👍
The quality of this tweet keeps going up. I remember many people DM me saying I can’t say this, but I insisted not to delete 😅
Shout out to @fede_intern for calling it out the first……my bar for calling a project scam is usually high and I don’t like to fight on CT, but….😅
Shout out to @fede_intern for calling it out the first……my bar for calling a project scam is usually high and I don’t like to fight on CT, but….😅
Connectez-vous pour découvrir d’autres contenus