Obviously Eli is legendary, but that's the wrong take. FHE is hands down the best solution for privacy, and it's not even close. ** Let me tell you why. ** Eli's making a point regarding scaling/integrity, which is clearly ZK's strong suit. But he's completely sweeping under the rug the fact that ZK is not suited for privacy. Yes, ZK is not meant for privacy, at least not in a meaningful, programmable way: 1. ZK only gives you 'single-user' privacy. Meaning, each user can compute over their own data privately, but any logic that requires combining multiple users data together, without revealing them to each other - is dead on arrival. The canonical example I like to give here is sealed-bid auctions. ZK can't even compute that simple task --> there's a large class of private computations that ZK simply cannot achieve. 2. There's an ugly hack around it - server-side ZK. Basically, it means that you have a single server (e.g., a zkEVM operator) that receives everyone's data in the clear (e.g., everyone's bids), then - it runs the computation over the plaintexts and proves externally to all others that it did so correctly. Because it's done in ZK, no one else learns the details. But Eli is talking about centralization risks of FHE, while ignoring that using server-side ZK is so much more centralized and provides no privacy guarantees at all. There's -full trust- in some unique operator. FHE solves all that. Everyone's data is encrypted, and anyone can compute over that encrypted data. Simple. No problems. ------------------------ So now that we've established that ZK is very limited in terms of privacy, let me address Eli's point regarding FHE-vs-ZK scalability. Eli rightfully says that FHE itself is untrusted, and I will concede that proving ZK-over-FHE is still underdeveloped and way too slow. However, because FHE is public transcript, you can basically just use optimistic proofs (a-la @arbitrum) to prove correctness. The market has shown that optimistic proofs are good enough for the time being, with @arbitrum and @base getting more usability than all zk-rollups combined. If it's good enough for plaintext, it's good enough for FHE. Also, remember that ZK had a pretty significant head-start. Companies like @fhenix and Zama are much more nascent and not nearly as well-funded. Give FHE 5 years and a billion dollars and you'll see zkFHE becoming a reality. TL;DR - FHE is the end game for privacy, ZK is for scalability. Since we're now playing the privacy arc, it's time for FHE to shine.