One way to improve the situation around selfish voting is to make the computation of rewards cyclical, and apply to the total curation reward for the cycle a weight that grows super-linearly as a function of the total number of votes, exactly like it's already done to encourage concentrating stake on the same account. That way, instead of concentrating their firepower to vote on only a few dozen "bluechip" authors, bots will have an incentive to spread voting large and wide to increase the reward weight. This can also be implemented without a fixed cycle: the system can keep paying the normal reward as post payouts happen (without bonus), and pay a bonus once daily based on the number of posts that have been upvoted in the last 24h and the total payout received for these posts.