On the one hand, the latter avoid the weaknesses of the former, in that they can neither easily be 'taken over' by outside forces nor provide a single point of weakness for their enemies to target.
And that's useful for both software and protest movements, since it leave a single target for the opposition to go after, whether that's through legal threats, harassment, arrest or assault/murder.
At the same time, it also makes the system prone to abuse by bad actors, since there's no one able to either get rid of them or distance the group from them in general.
Hence the extremists taking over political movements and protests, and the trolls/extremists/criminals becoming an increasingly prominent part of any decentralised community or social network setup online.
This in turn leaves both open to attack from their opponents and the media, who can point at the worst elements and say "See? These people are all evil/sociopathic/crazy/whatever' based on it.
It's like to some degree, avoiding bad actors and running a robust and uncensorable system are fundamentally incompatible with one another, in the same way as usability and security are.
In situations where mob mentality can take over and chaos can create a "fog of war" this is not true! Even though most of the protesters want to remain peaceful, the amplitude of sowing chaos, and its potential to spark additional chaos within mass rally situations, are providing extremists that "single point of weakness."
Basically, against an organized opponent that wants to create chaos, mass rallies are kind of like packing together certain isotopes of uranium atoms. Even if most of the units are "inert" (in that most of the people want peace,) if enough individuals with activation potential are densely packed enough, you can still get the chain reaction.