There's nothing wrong with centralized services built on a decentralized network. Take a look at the web. Sure you can use a centralized service like facebook to make a facebook page, but if you want you can host your own website.
It is a systemic failure that most users must "fail over" to a centralized service to publish on the web.
The conceptual improvement enabled by blockchain is that the data layer is a neutral plane and this theoretically gives users portability. But, to say that centralization is fine because it has happened on the web and that was also fine is rationalizing a bad thing as good actually.
sending http and json over the internet is just as neutral of a technology as the blockchain. the reason people build centralized services on top of it is that we're collectively better off by specializing.
As a writer you're better off writing your content full-time than running a server, becoming a smart contract expert, casual coder and server administrator. no technology on earth is going to change that fact and it's why people buy their bitcoin on coinbase and their nfts on opensea.
Specialization doesn't come into it, although it is worth observing: your comment presupposes that in order to benefit from specialization, one must subject themself to exploitation.
they don't stay there because they are being exploited, it's because centralized platforms reach many customers and automate away a lot of the problems that you'd have if you had to set that infrastructure up yourself and pay for it.
If you're a creator you stay on Youtube because you get a sustainable income. They can do that because they're centralized. If you run your own peertube instance and pay more in server costs than you make in ads that isn't a solution.
And these simple economies of scale will always exist regardless whether your software runs on a server or a distributed blockchain vm.
But, the commenter I take issue with said that it's good actually that we are just re-inventing banks on top of the blockchains, and the evidence that they offered is that it was good actually that we invented Facebook on top of the decentralized web.
Actually, Facebook is bad. It's not bad because it offers a service that people clearly need. I think we can all agree that an easy publishing on-ramp is great, and that it should exist. Facebook is bad because its creators use it as a tool for mass manipulation and exploitation of its users, and the web as we know it facilitates this through a combination of its flaws and missing features.
Facebook has created some useful services, but that doesn't mean we want to build Facebook again on a newly conceived network infrastructure. In web3 we see a latent opportunity to design new infrastructure that improves on the flaws and power imbalances of the previous one. But, if we seek only to re-invent the exploitative institutions of the past atop a new substrate, then frankly it's a waste of (most) people's time (I'm sure some VCs and early adopters have a lot riding on it, though).