Re:10,000 Mechanical Turkers work separately to draw a 100 d...
by serhei on Wednesday September 24, 2008 @10:29AM
(#313514)
If you scroll down you can see that some people drew completely unrelated things. One guy doodled a stick figure. Another wrote "$0.01!! REALLY?!" The last guy did the drawing, then wrote "i hate u" over top.
Re:10,000 Mechanical Turkers work separately to draw a 100 d...
by danw on Wednesday September 24, 2008 @11:49AM
(#313631)
Paying $0.02 instead of $0.01 for a hit gets you faster response and better quality usually
Re:10,000 Mechanical Turkers work separately to draw a 100 d...
by dcurtis on Thursday September 25, 2008 @01:02AM
(#314535)
I wonder if you paid two people 1c and then merged the results, how accurate you'd be.
Re:10,000 Mechanical Turkers work separately to draw a 100 d...
by danw on Thursday September 25, 2008 @04:56AM
(#314686)
You still wouldn't get the speed benefits that $0.02 offers. Some tasks (like this drawing one) can't be automatically verified.
Anyone compared 2c and 1c with doublechecking for simpler tasks?
Re:10,000 Mechanical Turkers work separately to draw a 100 d...
by shafqat on Wednesday September 24, 2008 @10:33AM
(#313520)
That kind of stuff is pretty rare. Turkers are pretty hard-working/honest people on the whole I guess out of 10,000 maybe 10-20 screwed around. That's less than .2%.
Re:10,000 Mechanical Turkers work separately to draw a 100 d...
by timr on Wednesday September 24, 2008 @01:42PM
(#313801)
They're certainly harder working than I'd be, given the amount of effort for the return.
Re:10,000 Mechanical Turkers work separately to draw a 100 d...
by serhei on Wednesday September 24, 2008 @08:24PM
(#314277)
Obviously the dollar bill guys picked the worst ones because they wanted to make their point about crowdsourcing. Whatever that point might be.
In any case, it's more interesting to stare at than just 8 well-done samples.
Re:10,000 Mechanical Turkers work separately to draw a 100 d...
by rudyfink on Wednesday September 24, 2008 @05:03PM
(#314077)
Looking at the bill, the number of bad cells seems far higher than you suggest. Some people didn't absolutely screw around but many did not finish the work. The percentage gets pretty bad in some of the high detail areas.
Re:10,000 Mechanical Turkers work separately to draw a 100 d...
by shafqat on Wednesday September 24, 2008 @10:28AM
(#313512)
I'm a huge fan of MT, and do think that it can be used to get pretty accurate/high quality results. You can actually have multiple people work on each HIT and then cross-check or chose the best
I actually posted about our experience with MY yesterday here: http://blog.newscred.com/?p=157
Re:10,000 Mechanical Turkers work separately to draw a 100 d...
by JulianMorrison on Wednesday September 24, 2008 @11:50AM
(#313633)
An algorithm you could use: try three times, prefer the most similar two ("map"), look in those for puzzle-piece matches with the nearest neighbors ("reduce").
Re:10,000 Mechanical Turkers work separately to draw a 100 d...
by jyothi on Thursday September 25, 2008 @06:23AM
(#314723)
http://www.tenthousandcents.com/gallery.html
This is the most interesting piece. You are seriously missing something if you are not checking this out. There are multiple sections -
Detail (Mind blowing! - people putting in so much into this)
Creativity - All the funny stuff
Random art - hearts, texts and all that.
Guess the Mechanical Turk isn't good for, say, important data entry (without a system to set up cross-checks).
Pretty cool, though.