4 ms·

You can certainly see who the lazies are. Especially prominent: the "H" in "THIS", the "LE" in "LEGAL", part of the crest, and part of the seal. Guess the Mech

You can certainly see who the lazies are. Especially prominent: the "H" in "THIS", the "LE" in "LEGAL", part of the crest, and part of the seal. Guess the Mechanical Turk isn't good for, say, important data entry (without a system to set up cross-checks). Pretty cool, though.

shafqat 18y ago
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 http://blog.newscred.com/?p=157
- JulianMorrison 18y ago
  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").
serhei 18y ago
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.
- shafqat 18y ago
  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%.
  - timr 18y ago
    They're certainly harder working than I'd be, given the amount of effort for the return.
  - rudyfink 18y ago
    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.
  - serhei 18y ago
    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.
- danw 18y ago
  Paying $0.02 instead of $0.01 for a hit gets you faster response and better quality usually
  - dcurtis 18y ago
    I wonder if you paid two people 1c and then merged the results, how accurate you'd be.
    - danw 18y ago
      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?
jyothi 18y ago
http://www.tenthousandcents.com/gallery.html 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.

shafqat 18y ago

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 http://blog.newscred.com/?p=157

JulianMorrison 18y ago
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").

serhei 18y ago

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.

shafqat 18y ago
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%.
- timr 18y ago
  They're certainly harder working than I'd be, given the amount of effort for the return.
- rudyfink 18y ago
  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.
- serhei 18y ago
  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.
danw 18y ago
Paying $0.02 instead of $0.01 for a hit gets you faster response and better quality usually
- dcurtis 18y ago
  I wonder if you paid two people 1c and then merged the results, how accurate you'd be.
  - danw 18y ago
    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?

jyothi 18y ago

http://www.tenthousandcents.com/gallery.html 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.