The all-purpose UNU thread...

Hey all, this is GammyLeg from UNU. We've hopefully crossed paths on UNU in the past...or hopefully will soon in the future! There are a couple of us lurking about here, so we'll try to use this thread to keep you all posted on UNU stuff we're working on or have coming up.

I can't promise we'll reopen the EPL qual soon, but look for us to have some things coming with other sports through the rest of winter. Like last year, we'll also be doing some swarms around the big awards shows coming up.

 

I keep meaning to ask you...is gammy leg a reference to Dr. Strangelove, Monty Python, or something else?

 

It's a Monty Python thing. I'd been using throwaway phrases as usernames to test, and we got to the end of that test....and I was sorta stuck with "GammyLeg". Ah well.

 

So for the Grammy HIT we published earlier tonight:

1. That swarm is at 11 am ET. I'll put a reminder in the HIT thread tomorrow when we're close.
2. We really would love it if you'd do a little snooping around on this one. You can go listen to songs if you want, but even better, look to see what some of the music media are predicting for some of those big awards.
3. And that's why we're offering a decent bonus for this. It's totally winnable, perhaps, if you're already knowledgeable, or can bring some new knowledge to our swarm tomorrow.

Good luck!

 

All bonuses from the Oscars and Daytona 500 should be paid out. Congratulations to the winners!

(BTW, one correct answer threshold out of 16 got you in the money with one Oscars swarm, but not in another...)

Overall, our Oscars Swarms were great! If Denzel had won instead of Casey Affleck (definitely the closest survey vote we saw), it would have made it a little better. In any event though, thanks so much for your help on that project everyone!

 

OK folks, tomorrow is the start of one of the biggest projects we've ever done at Unanimous AI--and it's all March Madness related!

Sometime tomorrow night after Selection Sunday, we'll post a $2.00 HIT. It will contain instructions for joining a special Unanimous AI bracket contest. That's right, we're going to pay you to do a bracket and compete with a bunch of other folks! We're also going to pay out yuuuuuge bonuses for the winner as well as second and third place. Like, bigger bonuses than we paid for the Super Bowl.

The process of signing up for a bracket is really pretty simple, but we've put in step by step directions for doing it, just to be sure. The most important parts of it are this:

Write down whatever USERNAME and PASSWORD you use to sign up with. Since we REALLY DO NOT want ANY personal information, you're going to use a string of nonsense characters for your email and won't get a confirmation reminder with that username and password. So...write it down, because YOU WILL NEED IT!

Second, make SURE your bracket's name is your Mechanical Turk Worker ID. If we can match a bracket to your submitted HIT, we can't pay you or make you eligible for the bonuses.

I will be around all night tomorrow and all day Monday if you have any problems. Don't panic, we'll sort it out and get you in on the fun! Post any questions you have in this thread, or PM me here. I'll get back to you ASAP.

Finally, this is separate from any bracket pick swarms we'll run next week. This is just you picking a bracket as a HIT and hopefully making some serious bank for doing that!

 

Hey everyone! I have a couple of things that I wanted to be able to let everyone know about in a place where we could link to it in case folks had questions.

1. If you've done our EPL hits this season so far, you've likely noticed we have a new format for deciding on a bet confidence on our predictions. It's a hex and then it drills down on a question automagically and immediately to a more specific answer. So....we like what we're seeing with these and so not only is it not going away, we're going to start using it in MLB for Friday and the res of the season.

What that means: our EPL swarms are a little longer, our MLB swarms take a little longer. It also means we're going to be paying more for those HITs! (Typically I figure roughly 1 minute per question; the new format doesn't take as long as a full extra question, but it's about 30 to 45 seconds per instance)

What that also means is that if you don't like that system, even if we're paying more....well, don't take those hits. For now anyway, we're using that system.

2. We are going to be rolling out, hopefully by next week, a new system that maybe makes the sports prediction hits a little more fun. Basically, we want everyone on the same page for what our "How much should we bet on that?" question means. We'll lay it all out, the short version is that we're going to track how much (or not!) the hypothetical bets you make on surveys, and whether the fake bets the swarm in a given week pay out. Instead of most wins in a week winning the weekly bonus, the person whose total fake bets paid the most wins. And we're upping the pot to $25 bonuses weekly. Because y'all are awesome. We'll also post a weekly google doc with the top ten standings so that if you think you did pretty well, you can check for yourself!

3. FINALLY...we are going to be collecting as much data on game predictions as we can get over the next few weeks. So, we'll run a swarm with a survey the way we've always done. But, we're also going to be running survey-only hits to collect a bunch of those predictions. So...if you miss a swarm, hang out, and grab the survey only part that follows.

Hit me up with any questions, and thanks again for all your outstanding work for us! Unanimous AI is growing and growing, and we're very grateful for the part our Turker community has played in that.

 

What's going on with the Current Events UNUs? Too much sports, just waiting for the New Year, waiting for the Yellow/Red card implementation, or ... ?

 

We're redrawing some tracking questions that will be valid over the course of at least three months or so, and also working out methodology so that we can get an even mix so that if swarm answers fluctuate wildly week over week, it happens in reaction to real life events, and not from the ideological makeup of the turkers who show up.

 

What's the algorithm for "points" on the NFL picks prize?

Depending on how complicated it is, it might be really cool to have access to something like a shared or "public" Google Spreadsheet (or Doc) that shows how each entrant did and where they did well or poorly at least for the group each week, or even more fantastic would be a sheet for each entrant over the season (understanding that people probably own't be able to participate every week for whatever reasons) where the focus is on just their data. Depending on how you're crunching the data some of this might be easy, and some might be an understandable "oh hell no."

Do you feel the need to keep the points algorightm static all year long?

Did adding another? prize more related to picking against the spread get discussed?

 

So starting with tomorrow, we're going to begin to experiment with this method for paying out bonuses on EPL swarm HITs.
Take a look, folks, and let me know if this wording makes sense, and if it's understandable and clear.

---------------------------------------------------------------------------------------------------------------------------------------------------------------------------

(PLEASE READ! This will be our scoring system for paying the bonus for this week.)

For each of our game predictions, we’ll ask you how much you’d like to bet on your own pick, from $0 up to $100.

We’d like you to imagine that you’ll be given UP TO $100 in fake money to bet on EACH individual game prediction.

You can bet up to a hypothetical $100 on each game, but you don’t get to keep or “bank” any fake money you don’t bet on an individual game.

You earn points equal to the amount of your bet if your prediction on a game is correct.

You lose points equal to the amount of your bet if your prediction on a game is incorrect.

When all games on this survey are complete, we’ll total your win amount, and subtract your loss amount.

The person who earns the $25 prize each week will be the person with the highest total hypothetical winnings across all games.

IF THE SWARM FINISHES IN THE TOP TEN FOR ANY GIVEN WEEK, WE'LL ADD $5 TO THE BONUS TO BE PAID!

IMPORTANT:

1. You can bet between $0-$100 in hypothetical money on EACH game. Theoretically, you can bet $0 on each game, and also bet $100 on each game. (In our testing, neither of these are good strategies. "Bet" based on your confidence.)

2. Money that you don’t bet doesn’t count towards your total earned or lost.

3. Your total score each week is your amount earned from correct predictions minus the amount lost from incorrect predictions. If your score is the highest, you win the $25 prize.

EXAMPLE:

You make four game predictions in a week, and three of your predictions are correct, one is incorrect. You wager $50, $40, and $30 respectively on your three correct predictions (plus $120 total) and wager $20 on your incorrect one (minus $20). Your score for the week is 100. If that ends up being the high score, you win the $25 weekly prize. If the Swarm's score finishes in the top 10, you'd win $30.

Can I check my score to see how I'm doing?

Yes! We'll post and update scores (by UNU usernames) at this link this week. (You may wish to bookmark that.)

We'll also keep the swarm's score at that link. Remember, if the swarm finishes in the top ten, we'll add $5 to the $25 prize for the week.

 

Cool. Nothing like potential public humiliation to focus the mind

Bookmarking a link to the latest results would be preferable to bookmarking to a specific week.

Which begs the question ... will the data stick around or will it be replaced in this spreadsheet each week?

A graph of the winning totals through the weeks would be useful information. Data dumps work too, but graphs are cooler

My brain is exploding trying to figure how to "bet" on those 28% of the time ties.

So the most confident tie and win don't figure into this calculation, right?

 

@MyLegIsFine

"what percentage of the people ... will pick the same WINNER you did?"

there are 5 choices, 1 of which is a tie.

Is my bet on the one of the 5 choices I made?

"winner" is very ambiguous here ...

 

Only if you're really overthinking it.

It is a little ambiguous, but the context seems reasonably clear. And worth noting: this is the same exact wording to that question that we use in our NHL and MLB swarms, with the only difference being the presence of a tie game choice.

For now, there's a reason we have that standardized. In our tests, the context seems clear enough, but once we're able to adjust the wording we may do just that.

 

So we've completed our test of our wagering game from week 1, and we're pretty happy with how things turned out!

To be clear, the weekend in the EPL was totally bizarre and volatile. There were 5 ties. We're *never* going to perform well as a swarm when there are more than 2-3 ties in a given weekend, and we kind of are at the acceptance point on that.

And then there was whatever the hell happened with Chelsea and Crystal Palace. Crystal hasn't scored a goal all season until that match. 731 minutes of Premier League play, no goals. Then they score two, and win.

So yeah. Sometimes you have to shrug and realize that sports are going to do sports things.

But here's the interesting bits. Of 36 participants, our swarm finished tied for 27th for number of games picked.

That's bad. (That's the worst its ever performed on that metric, in fact.)

But...the swarm finished 14th in wagering totals earned/lost.

We're eager to try this again this weekend for EPL (hopefully on a weekend when pigs don't fly or other weird stuff happens), and then introduce it going forward for all sports that we swarm.

One change we'll introduce: instead of paying the winner an extra $5 bonus if the swarm finishes in the top 5, we'll bonus the top ten participants $1 each if the swarm hits the top five in a given week.

 

@MyLegIsFine

Hi!

How do we get the financial qual?

 

It was part of a qual hit we put up on Wednesday.

 

aww I missed it then, thanks!

 

Hey everyone! Later this week we're planning to do our own Bowl Game predictions. We're not covering all of the bowls the way @Hummingbirdee's bowl pick 'em does, but we will cover the biggies.

We'll be paying a $50 prize to the high scorer, $20 for second, and $10 for third. If you want to get a jump on researching the games we'll predict, here they are:

Camping World Bowl: 22 Virginia Tech vs 19 Oklahoma State
Alamo Bowl: 13 Stanford vs 15 TCU
Holiday Bowl: 18 Washington State vs 16 Michigan State

Cotton Bowl: 8 USC vs 5 Ohio State

Fiesta Bowl: 11 Washington vs 9 Penn State
Orange Bowl: 10 Miami vs 6 Wisconsin

Peach Bowl 12 Central Florida vs 7 Auburn
Citrus Bowl: 14 Notre Dame vs 17 LSU

Rose Bowl (playoff) 3 Georgia vs 2 Oklahoma
Sugar Bowl (playoff) 4 Alabama vs 1 Clemson

We'll also do some hypotheticals with the championship game based on the final four here.

 

Remember last year, when everyone in the swarm mocked me for picking Clemson?



Also, this made me feel famous. Yes! Birdie Bowl! Everyone knows about the best bowl pick em!