Voting Bot Update
It's been a weird week for the voting bot.
image source: pixabay
Recently, I've attracted a good many vote-followers (or, as @liberosist calls them, "stalkers") who blindly follow my vote. This is kinda neat! Now, if I and my clients vote for a post with full voting power, its potential payout immediately jumps well above $1; see for example this @son-of-satire post that has a potential payout of $1.74 right after I and my train voted for it.
Nominally, this would be good for my curation rewards. But it turns out, as @shenanigator pointed out to me last week, it comes with a significant downside: @curie only votes for posts that have less than $1 payout. So apparently that means Curie will no longer be voting for stuff that I've voted on! And, like everything, even this is both good and bad. Good, because it means that I'll no longer be leaching off Curie; bad, because Curie's votes (on posts that I'd already voted on) used to be a major source of curation rewards for me. So what could happen is a self-defeating feedback loop: my bot predicts that a post will make a good payout because it's something Curie might like, but then this becomes a self-defeating prophecy whereby my vote causes Curie not to vote for the post. My adaptive payout-predictor model will learn about this, but the net effect will almost certainly be to reduce the accuracy of my model.
So we'll see where this goes. Everything in the curation game should be more-or-less self-correcting: if I really need Curie to make good rewards, then without their votes my curation efficiency will drop, my clients will leave me, and my voting train will dwindle to the point where I'm not voting stuff over $1 anymore, Curie will vote after me again, and I'll be back in the money. The curation game in Steem is full of these sorts of negative feedback loops, and one of these days I'm going to write a post all about them.
But anyway, I haven't told you the algorithm update yet: I just did some analysis on my bot's adaptive algorithm, and I discovered that my old payout-predictor model has gotten far less accurate over the past month. This is certainly connected to the dwindling rewards, but it's not completely obvious which caused which. The point is, I've switched prediction models to an alternative one that I've been tracking lately that has been consistently outperforming my old model. This won't make my rewards stop dwindling immediately, but it should help. In the long run, I'll need to adopt a more robust solution.
Bot world problem.
Something else worth noting - earlier this week the magic number for Curie dropped from $1.00 to $0.50.
This whole situation makes for an interesting dynamic. Yours was the most profitable bot, so it gained a lot of "stalkers." Now that your train is too powerful for Curie votes, someone else's will be the most profitable. I suspect the same thing that happened to you will also happen to them.
$0.50 is a great move by Curie. They'll be able to earn vastly more curation rewards, which will help fund their operations. It also increases the challenge level for curators, which livens things up for all of us.
This is super interesting. If Curie's submission filter's entirely automated, do you think a $0.50 threshold would've been suggested and maintained by the algorithm? Just wondering if a bot could arrive to such a decision that easily.
It's totally impossible to say, but that would be a fun problem to work on. It would depend completely on how you defined your objectives. We should talk about this in about 6 weeks, after some of my deadlines have passed and I have a little more time to do stuff. :)
This post has been ranked within the top 25 most undervalued posts in the second half of Feb 10. We estimate that this post is undervalued by $11.14 as compared to a scenario in which every voter had an equal say.
See the full rankings and details in The Daily Tribune: Feb 10 - Part II. You can also read about some of our methodology, data analysis and technical details in our initial post.
If you are the author and would prefer not to receive these comments, simply reply "Stop" to this comment.
Interesting...
Welcome back to mortality, @biophil!
Heh. It's but a flesh wound.
We'll call it a draw?
try voting at 99% instead! and badabing, badabang, problem solved, that easy, just like that (maybe)! 😁
You would have hoped it'd be that easy, but alas no. In fact, I had voted for the post I linked to above at a rate of something like 70%, and it still went to $1.74. A full 100% vote would have put it over $2!
Another thing is that I'm not wildly interested in sucking up all of Curie's curation rewards, so I don't particularly want to custom-build a bot for the purpose of front-running Curie.
I figured you probably tried it, but you hadn't mentioned it. So I figured I'd throw it out there at least to "see what happened" if/when you did lol
It's amazing how the bots and people continue challenging one another to improve and move forward. Nice of you to update us!
It's a fantastic set of interesting challenges. I've enjoyed the game!
It might still work out if the posts you are upvoting end up in the "hot" section, and normal members just browsing for good content see them and upvote them. It should expand the number of posts that get paid.
That is certainly my hope.
How will you know which posts to front-run Curie on? With $1 threshold, the search space becomes rather large.
I've never explicitly tried to front-run Curie, and I don't intend to start doing that. My learning algorithm essentially got pretty good at front-running, but now that's effectively over. It wouldn't be too hard to program it to vote with a low enough weight that the follower-train wouldn't put the payout over $1, but I don't have time to work that up right now.
I am confused ! Totally puzzled about this world of mortal bots
I don't think I understand your question. My bot might occasionally vote for you, yes.
Sorry ! You replied before I could edit . I was just confused and writing on my phone is not as easy either. Yes I wanted to know how you choose and who you vote for ?
I programmed my bot to learn which posts are going to make high curation rewards. Then it votes for some of those posts. That's about it; I have a little more information about it here: https://steemit.com/curation/@biophil/how-my-bot-learned-korean
I also have it vote automatically for several other accounts on Steemit, and some of them pay me a share of their curation rewards.