Milestone reached! I'm the best curator on Steemit.

in #curation8 years ago

I submit as evidence the following screenshot of the Steemwhales.com 28-day trending curators rankings:



Yes, that's me on top of the list. I just passed @laonie (the previous automated-voting champion) on this list yesterday. To what do I owe my success? Well, in case you haven't been paying attention to my posts lately, I run a voting bot. It has been voting pretty successfully for a month now, even for languages I don't speak. I've been extremely happy with its performance, and I've been gradually creeping up on @laonie over the past weeks.

How does this help you?

Because you can vote with my bot! Last week, I invited the community to join my voting bot and start earning some of its curation rewards. Since then, at last count, my bot has earned people in this community over $115-worth of SP (at today's prices). For reports on how much it's earning whom, follow my other account @ozymandias. I post updates now and then over on his blog that show curation reward totals; I may post other interesting metrics as I develop them. Here's a recent one: Curation Report for 2016-10-26 22:57:21 to 2016-11-23 21:46:54.

What sorts of posts does the bot vote for?

My bot is different from many, in that it's content-unaware. It simply learns what sorts of posts earn large rewards, and votes strategically for those. Because of this, it almost never votes for things that make it to the front page. Most of those posts have all their curation rewards used up early, and are rather difficult to vote for profitably in an automated way. On the other hand, if you can find a post that is likely to have a $10-20 payout, and vote for it right at the 30-minute mark, you can earn significant rewards. This is my bot's strategy.

How can people in the community join the bot and earn rewards?

If you want me to add you to the voting bot, just send me your Posting Key and account name on steemit.chat or to my email address that you can pull from my website. I'll add you to my bot and you'll start climbing up the trending curators rankings as well.

Does it cost anything?

Not yet. Some of my clients are tipping me, which I appreciate very much. I always vote my account first on every post, which means that I earn slightly higher curation rewards than my clients; this suffices as a payment for now. I'm contemplating implementing a bidding system where my clients bid on vote position, but I haven't designed the auction yet. So for now, it's free. If you're a very small account, I reserve the right to delay your vote to a bit later than my bigger clients.

Is it ethical to use a bot for voting?

That's a hotly-debated question, and as a bot creator I suppose I've forced myself into that debate. Of course you'll expect me to think it's totally ethically fine; but all along I've been saying that serious trouble could come if bots end up being too prolific. My bot chases profits - it's purely greedy. I have a few accounts blacklisted because I don't believe in their message (example: fear-profiteers like @dollarvigilante), but for the most part my bot just votes for what it thinks will earn it rewards. This is fine in principle, but if I get too many voters following my bot, then my bot's votes will actually start affecting the profitability of posts. It will learn this, and each of its votes will end up being a tiny little self-fulfilling prophecy. This type of singularity seems like it would be bad for Steem.

Now, I think the incentives are structured such that having a huge pile of vote-followers would not actually be rational; but one of the long-standing problems with Steem is that most voters don't know a single thing about how curation rewards work. Incentives are only valuable if people understand them, and the dev team's apparent hatred of documentation works against this.

So, I think automated curation is intriguing; but whether or not it is helpful is probably completely dependent on the incentives.

Sort:  

This post has been ranked within the top 50 most undervalued posts in the second half of Nov 27. We estimate that this post is undervalued by $5.95 as compared to a scenario in which every voter had an equal say.

See the full rankings and details in The Daily Tribune: Nov 27 - 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.

What's your secret?

My secret is my learning algorithm. It looks at what every steemit post earns, and figures out what features of each post can be used to predict its success (or lack thereof). Someday I'll probably publish my algorithm, but for now it's my secret.

This is intense. I love it. Also I didn't knew @ozymandias was your account. I followed it's votes a couple of times.

Hi, it looks like other top curators are using quite diverse strategies.
@laonie usually have voting power around 50+%, on the other hand @mata have voting power around 90+%.
Your bot have 60+%, have you done some experiments with intensity of voting, or it just happens so ?

Voting power is one of those thing that the devs have explained very badly, so many people think that there's something bad about having low voting power. The reality is that voting power doesn't matter at all. I could drop my bot to voting at 75% on each vote, and my account's power would climb to around 100%, but I'd get exactly the same results after it equilibrates. I like my voting power to be a bit lower, because then I have a buffer if I have to shut the bot down for a while for some reason.

I mean, @mata's strategy seem to be to cast not many votes ( but mostly 100% weight) this keeping total voting power mostly over 95%. On the other hand there's @laonie making a lot of votes, often with only 30-60% weight, but because of many votes his total voting power is often just above 50%.
That seem to be very different strategies, but at the and both are successful.
Btw, did you got an email from me ?

Yeah, I did get an email from you. Sorry I never replied.

I see what you're asking. I haven't really tried to analyze @laonie or @mata, but they're both quite competitive. For a while a few weeks ago, it looked like @mata was the bot to beat, and of course @laonie was the champion for a long long time.

The "how often should you vote" question all comes down to a balancing act. On the inside of my bot is an adaptive model that predicts each post's payout; the correlation between actual payouts and predicted payouts is not great. Here's a scatterplot of the past week (horizontal is my prediction, vertical is actual):

They're obviously correlated (the correlation coefficient is around 0.65), but I can't predict with 100% accuracy how well a post is going to do. So to compensate for that error, I have to spread out my votes over more posts.

If my model were perfect, I'd vote for the best 40 or 50 posts each day and be done. Since my model isn't perfect, I vote for many more. I haven't done nearly as much validation as I'd like to see if I'm actually voting optimally given my model; I have a lot on my plate right now and I just don't have the time. When someone comes along and starts consistently beating me, I'll put some more work into improving the algorithm. :)

Thanks for writing this post. It's the first time I have been reading about voting bots (I knew they existed, but never really looked into it), and reading your past posts has been interesting. I sent you an e-mail with the posting key, so I'm really exited to see how this works :)

You're in! Just to warn you, your curation rewards will be pretty low because your steem power is low. You will earn, but it will probably be less than 1 SP per week. Despite that, your account probably will show up on the trending curators on Steemwhales.com in a few days :)

Thank you! Yea, I figured it would be low, but I'm sure it will improve once I manage to post some more cool content that people like :)

You can see what posts your account is voting for at https://steemd.com/@valth. Also, you'll still be able to vote for anything you want - and if you see the bot voted for something you don't like, you can un-vote freely.

I got a kick out of this:

Incentives are only valuable if people understand them, and the dev team's apparent hatred of documentation works against this.

If we had a line-item up-vote, that one would certainly be a hit. ; -)

....in that it's content-unaware. It simply learns what sorts of posts earn large rewards,

hmmmm.

Well congratulations anyway.
Not everyone can do what your 'bot did.

Hi @biophil, I sent you an e-mail with my posting key

yep, you're active.

Hi sent you an email to join. Thanks again for opportunity.

hey sir.
it will be hard to battle you and your crew.
this is really exiting !
curation with bots in competition will be an extremly interesting and valuable experiment for the future.
The results of these experiments will be very instructive and useful for upcoming usecases of bots.

Yeah, I like the sound of it. It could spur some really interesting things!

Great post, You explained this well. I like the fact that your bot has a type of randomness to it , rather than just voting on every post that one profile makes.

Nice job !!
Followed. Upvoted.
Follow me back guys, I will do the same 😘

Your upvote doesn't do anything to posts that are more than 7 days old. How did you even find this post?

nice post , but your bot still work ?

Yep, still voting away!

I have this questions, @biophil so after the 7 days mark what happens to the extra upvotes on a post

Nothing. They are just for show.

i was looking for some information about curation and bots.
i am new, at steemit comunity, but is yet the posiibility to be included in the bot? Thank you @biophil

Welcome to Steemit! Let me know when you have gotten to about 1000 Steem Power, and I can add you to my bot. Until you have about that much, my bot isn't going to make you much money. Good luck!

Ok! Thank you so much! I will contact you again when whe achieve the 1000 SP. Thanks again! ;)

Do you have the code published somewhere ?

No. That would make it too easy for someone else to exploit the algorithm.

:-)

May be make better too ?