Sybil attacks do not work in small communities that members may choose to form where the members already know each other.
If a system forces all users to be part of some large, Borg-like, distributed hash table, or ledger, then by my definition it's not "fully decentralized".
Indeed, if you don't plan on writing distributed systems that work for more than a few people, you don't need to worry about Sybil attacks. However, the nice thing about the internet is that it connects billions of people, so here we are.
I think there's a lot of historical evidence over the last few thousand years that people naturally form small communities, or at least small groups within large communities.
Today, people can, in theory, choose from among billions of peers to form these small groups. And the groups can if they so choose connect with each other, via a network of networks.
This internet "connects billions of people". True. But your company's LAN probably does not connect that many.
If a user started creating numerous fake identities on the LAN, then it's likely she would be detected.
Is it possible to create distributed "LANs" over the internet?
(rhetorical question)
Another commenter questioned why a distributed Web needs "lack of trust".
People in small groups can and do trust each other. No computers are needed to make this happen.
Fortunately the two approaches are not mutually exclusive. There are no rules about how the "distributed Web" must be constructed. As the old saying goes, there's more than one way to do it.
If a system forces all users to be part of some large, Borg-like, distributed hash table, or ledger, then by my definition it's not "fully decentralized".