Ok, answering my own question:
Found a table that cited the effects on the weights here. This table was initially surprising to me, but actually it's missing the higher tiers that control pretty much the entire rewards pool (though I'm not even sure what the distribution looked like back then). But at least now I see the point people were trying to make: actually makes it so pretty much only one person can self-vote (the fellow at the very top).
So it appears that there is something I missed in terms of discussing power and self-voting, being highly dependent on the distribution, but still, it is wholly dependent on who on the top is on top exercising their high SP votes, and that would still be of concern to me.
But it boils down to the following: The real reason people are claiming self-voting power is reduced is because the total number of people that can self vote effectively is reduced to just the biggest whales (top 10 lol....). But this is still pretty concerning to me. Here's what it sounds like to me:
Before: Nobody can self vote, look at that. (But there's like a handful of posts getting all the rewards, what the heck?)
After: Oh look, we have more even distribution, but there are more self-voters.
The diversity of content being rewarded has increased, has it not? Is it, or is it not an improvement over the old system? Self-voters is a problem, but concentration is a large problem too, and I now see why superlinear does not work when the existing distribution of wealth is how it is.
I think I just summarized the arguments on both sides. Pick your poison, I guess. If you ask me, I prefer linear.
Not that different in broad terms.
Not really. This is a simplistic analysis that only considers individual self-votes. People are creative in coming up with games, businesses, services, explicit or implicit vote trading agreements (two or more semi-whales agree to each vote for their other's posts, etc.), vote trails with profit sharing, etc. in order to assemble up the necessary rewards to exploit superlinear. It was already happening during the n^2 era and would certainly increase over time.
Pretty much the only thing that superlinear is good for is reducing the absolute number of content items with payout so it is easier to find the undeserving and downvote them. (But, as I've noted elsewhere, this is useless if downvotes are too expensive.) It comes with a heavy cost in terms of making literal self-voting more profitable for those at the top and exploitative team voting schemes more profitable for the rest. Interestingly, the 'mild' superlinear that some propose, while mitigating the degree of rewards monopolized by the very top (or by large schemes), would at the same time increase the number of stakeholders for whom literal self-voting is more profitable (and similarly reduce the necessary scale of team voting to achieve profitability). No free lunch here.
Thanks, this makes a lot of sense. One thing I would venture a guess to, and do correct me if I'm wrong, is that if a group of people pooled SP into one account and had a comfortable SP lead, they could actually self vote and even if everyone downvoted it it wouldn't matter because their influence is that much more than everyone else put together? It's why I mention that distribution is important. The dynamics you mention seem like the case when multiple parties have roughly similar stake at the top?
Yes obviously if enough people pool their stake it could be more than all of the rest disagreed both in theory (at 50%+) and in practice (at some much lower threshold).
Whether it is multiple parties are not doesn't matter that much when people can come up with ways to pool, which they can.