It has become customary on Steemit to write a special post when you reach some meaningful threshold; the inspiration for my post today is that I've just hit 300 followers. Thanks all for liking my stuff enough to think it's worth putting on your feeds!
(image source)
I'm going to take this moment to share some thoughts I've had about Steem over the past few months of my activity, as well as my personal outlook and goals for the immediate future here.
My Story
When I first came to Steem, I thought it was a pretty fantastically-neat idea. For the past several years, I've been fascinated by Dan Larimer's various projects, and Steem piqued my interest. I love the idea of a Reddit-like site and actually pay contributors, and it seemed totally reasonable that a Larimer-style blockchain could provide a key piece of that.
As it happens, I also have a professional interest in Steem - I'm a researcher at UCSB in the area of incentives in socio-technical systems, and Steem is chock-full of fascinating incentive-design problems. So I immediately set to work on my series on the Game Theory of Steem. It was an almost-instant hit! I was lucky enough to get my biggest-ever payouts right in the middle of the huge Steem price bubble, so in my first two weeks on the platform, I earned something like $8000. It was beautiful. It was heady. I thought I had stumbled onto the most unbelievably-cool phenomenon of my life. People were paying me thousands of dollars to ramble about game theory!
Short-Lived Glory
Of course, there are very few people who can make a living rambling about game theory. I am apparently not yet one of them. After my third game theory article, the whale votes mysteriously disappeared. I wrote 4 more articles (four, five, six, seven) before finally realizing that the whales had more-or-less adopted a policy of not voting for content that was only about Steem. Since the money had dried up, I quit writing about game theory. This despite people telling me that they loved my work. Sorry followers, apparently I'm quite the mercenary.
Lost Content Digest
Near the end of my game theory series, I had this idea that I'd start a curation service where I'd go find articles that didn't pay out on their first payday and republish them. Then I'd send the SBD proceeds to the authors! I still think it's a decent idea, but it too never really took off; at least not enough to justify the huge amount of time I had to spend on it. But still, it got some people paid who otherwise wouldn't have been - and some of them are still around to this day. For instance, I see good posts from @lily-da-vine all the time, originally featured in Issue #2, which also happened to be the highest-paid LCD issue of all. So maybe it was a success after all?
A Normal Blogger
Game Theory wasn't paying, LCD took too much time, so I started simply writing articles. I put out one every few weeks, mostly about food I was making, houses I was building, and mountains I was climbing. It was good, the @curie project picked some of them up, I was earning a little bit again. But I had something interesting cooking...
My Adaptive Voting Bot
I had this one cooking for quite a while before I fired it up. When I was writing LCD, I used @xeroc's piston.steem Python library for a lot of automated tasks: finding candidate articles and generating the articles themselves. My goal, however, was to build a bot that learns how to vote for high-paying posts. Version 0.1 of that bot is now operational! In fact, many of you are following me because my bot voted for your posts - this was an unexpected effect that I'm grateful for. See how the follower count angles up when the bot is turned on:
I'm not publicly disclosing my learning algorithm yet, but it's a very simple reinforcement learning approach. The bot learns which posts make large payouts, and then votes for posts that resemble previous high-earners. I have a long way to go before it's a truly great bot, but I'm getting there. In fact, I'm currently (and temporarily) ranked #1 on the Steemwhales.com 4-day best curators:
It was really cool to see my account rise to the top so quickly, but it doesn't actually mean I'm the best curator. If you look at my profile on steemd.com, you'll see that my voting power is down near 20%. There's nothing wrong with this in principle (exc, but it means that since I turned the bot on, it's drawn a huge deficit from my voting power (effectively spending 9 days' worth of voting power in only 5 days) - and my initial success won't last. If you look at the 1-day rankings, you'll see that I'm way down the list already.
Watch out, @better! I'm coming for you! After I've gathered a week's worth of data, I'll make some significant changes that should dramatically increase my bot's efficiency. So keep an eye on that. By the way, if my bot really does become as successful as I believe it can be, I'll be looking for partners to help me monetize it. Hit me up on steemit.chat if you're interested!
The Future
I'm interested in providing automated services on the Steem blockchain. For example, I envision creating an army of bots that people can follow for various goals: perhaps one bot would simply try to maximize payouts; another might try to upvote poetry; another might try to upvote high-quality technical writing. I've yet to create the classifiers that would be required for all these, but it will be a fascinating challenge: as soon as I deploy a voting algorithm, there will be adversary algorithms out there trying to learn mine so that people can take advantage of my bot's votes. Thus, a key design goal of a voting algorithm is to be resistant to adversarial exploitation. If I can solve that problem, I'll not only make a lot of friends on Steemit, but I'll have something I can publish in the academic world. Two birds with one stone!
All in all, this has been an unbelievably fun experience, and it has given me opportunities to learn all sorts of new stuff, like how to program a bot to learn how to vote! Cheers, everybody! Keep on steeming!
nothing wrong with focusing on the material that pays out ;>
Game theory is one of those things I need to read up on, but the "to-read" list of mine seems to keep growing...maybe I'll be able to check out your application of it to Steemit.
Congrats!
I'm a BIG fan of your work @biophil - we are of similar reputational standing too. All the best to 500 followers next :))))
Well thanks! 500 here we come!
Disclaimer: I am just a bot trying to be helpful.
Weird. I guess you pulled that out of my image source?
Glad to be your 300th follower, keep on Steeming!!!!
And there you are! Now I'm going to edit the title... because it's just so much cleaner for it all to be 300...
I've learned a lot following you, and am interested to see how your bot improves over time. Having thought about it myself, I'd also like to know the steps you'll take to make the bot resistant to exploitation.
I'd like to know those steps also. :) As a first pass, I think randomness and adaptability are two extremely important pieces of it. Currently, my bot adapts. For example, right now it thinks the tag #curation is a strong predictor of large payouts. If many authors start spamming that tag for my votes, the value of that tag in my model will automatically decrease and my bot will stop voting for it. There will necessarily be some learning delay, but it should work.
Congrats on 300 followers and the success of your bot!
Congrats 300x. Thanks for upvoting my newest post just now. Up voted and already followed.
@biophil, am too late for this but Congratulations!
Congrats!
Congratulations! I'm very impressed with what you have achieved :)