pull down to refresh
I liked the log scaling simply because it represents more what a collective of stackers like and less how much sats individual stackers spend (and I thought that by removing trust from the equation, that this would become even more true, not less.) It decouples spending and earning from visibility a little, which is the opposite of money-is-the-moderator.
If I like your post very much under the current system and I zap you 10k, you're for most 4h windows the #1 post. Just from me liking it so much. But there are other posts that many more people liked, but didn't want to spend 10k sats on. Therefore, I think that the logarithmic system was a truer representation of what stackers liked, because the excesses got muted. Now we have whales deciding what is on lit, to the point that I no longer read it. Because I don't give 3 fucks about about coinjoins, yet half the front page being about it. Coinjoins are the most uninteresting part of bitcoin.
The inherent problem here is that there is no correct way to compare intersubjective utility, but we do want both intensity and extensity (surprisingly a real word) of post appreciation to be captured. The most agnostic way to do that is by just counting the net upzaps.
If trust were added back in, then your not caring about coinjoins would be accounted for.
The most agnostic way to do that is by just counting the net upzaps.
Agnostic to what though? In a closed-loop system I think that I'd agree with you, but the zaps are open loop, meaning that the agnosticism is to everything but external wealth? I.e. there is no measure of what is popular content, there is only a measure of how much externally acquired sats are put towards the promotion of the post.
Great for advertisements, terrible for a community.
It's not about open vs closed systems. The issue is that not only do neither of us know whether my love of something is greater than your hatred, it's a meaningless comparison. There are no units for such a comparison. What we have units for is the proxy to that of how much we're willing to pay to express our feelings.
We also don't know (and can't know) how well my vs your feelings map onto other people. If one in ten people love the same thing I do but four in ten marginally like the same thing you do, there's no non-arbitrary reason why the thing I love should be less visible than the thing you like. If it's more visible, then we're stacking the deck towards surfacing things that will be valued highly, if only by a few people, and if it's less visible, then we're stacking the deck towards things that will be mildly appreciated by many people. There's no right answer here.
Balancing intensity of feelings towards something against the consensus of feelings (I wish I had better terms for this) requires making arbitrary assumptions about how different people's feelings compare to each other. Giving everything the same weight (just counting sats) makes the fewest unknowable assumptions (sort of... or it's just as arbitrary. This is something econometricians argue about though.).
Question to this, as currently the only metric is sats spent, so we have this system of the fewest unknowable assumptions: do you think that the front page for say the past month, on average, was more valuable as a recommendation system than it was one year ago over the same month?
I'm asking because I think that if we want to measure what gets zapped the most, then measuring what gets zapped is per definition the best metric. And from a holistic p.o.v. this makes total sense. But from a usefulness perspective for the stackers that come and spend their sats here, as in head count not sat count, do we really think that sum(sats) is an actual good metric for quality content?
My cop out answer is that we don't know because we haven't approached the equilibrium of the current system yet, nor have we seen all the wallet-postponed improvements.
My straight answer is that I think Stacker News was in a better place in the past, but I'm not sure how much to attribute that to the ranking algo.
It's good to experiment with the system to see what each part is doing, but I always thought we should be using trust more rather than less and focus on improving the trust metric.
I'm not so sure I still favor the log scaling. I'd like to see what happens with a combination of trust and linear scale.