new time waster!

[piranha]

google image labeller, to aid googe image search. since i use that a huge amount, i am all for helping with it. you and a partner (picked for you at random) have a certain amount of time to type in labels for an image. you get points for matches.

edit1: *little grump*. i wish the handshaking weren't so buggy. i'm having too many passes not connect.

edit2: *wah*! and now you've taken my individual ranking away! how can i tell whether i am working my way up in the greatest hits list! you people have a lot to learn about incentives.

edit3: ah, i see what they are doing. they're presenting the images again and again, and are putting common terms off limits, thereby refining the labels.

edit4: i am refining my own labels by taking time after each run to see what my partner typed, and by going to sites for images where i drew a blank to check what they actually depict. otherwise i'd already be bored now.

edit5: my individual cumulative ranking is back. 302 with 14400 points. just had my first 10-match with somebody; that felt really good (and it wasn't just simple stuff either). i think this is good "flexing the brain" exercise for me.

edit6: if you cna't splel astronaught, DO NOT PLAY THIS GAME, or i shall have to kill you. 11-match, #234, 17600 points.

Comments

[akirlu.livejournal.com]

Wow, that has got to be a really interesting psychosocial experiment, whether it's being used that way or not. I find very different rhythms with different partners, and usually the way to get the most matches fastest is to provide the lcd response -- that is, you lose precision trying to get a label two people agree on, and interestingly it's often the more identifiable images that you can least get agreement on. Could not get anywhere with the photo of a combine harvester in a field by a tree. It's just really interesting to see how different the dynamic is with different partners. The most matches I seem able to get at one go is 9 although once I got the hang of it, I got several nines.

I do really wonder if this will actually improve the usefulness of the image search capability, though.

[pir-anha.livejournal.com]

yeah, i want to write several papers on it already! i can get a pretty consistent 7 without simplifying too much ("man", "woman", "people" are right out), but it seems hard to get more without losing way too much precision for me.

and agreed on the usefulness or lack thereof for the image results.

do you get a lot of CERN collider images? they're throwing them at me like there's no tomorrow.

[akirlu.livejournal.com]

do you get a lot of CERN collider images?

Oh is *that* what those are? Yes, I think so. Weird, high-tech-looking facility with lots of scaffolding and gantries and open catwalks and huge, radially symmetrical whatchacallits? If that's them, I'm getting lots of 'em. I mostly hit pass on those, though knowing it's CERN, I might try it. Or not. What I hate is when I just go blank -- I got a fractal/mandelbrot image and just could not think of either word.

[pir-anha.livejournal.com]

yup. it's the CERN LHC collider under construction. :) i looked it up, but have not come across somebody else who has looked it up as well. :) i am waiting for that strategy to start paying off; there are bound to be other geeks like us around.

oh yeah, i get the occasional "can't think of the word" blank -- *hate*!

[leiacat.livejournal.com]

Oh, _that_'s what it is! For me, it's been the Pass image.

[wiredferret.livejournal.com]

Yeah, after the third one, I went and looked it up. All the cern pictures seem to have a distinctive light color to them.

Once, my partner knew, too.

[leiacat.livejournal.com]

I've played with it a little, and found that it really pays to oversimplify. If I see a bulgarian stamp of Franklin, it pays me to tag it "stamp" because that's the lowest common denominator. My best run - 10 matches with the person - included 3 pictures labeled "guy", which is not going to be at all useful to anyone. I cringe at the results this generates.

[pir-anha.livejournal.com]

yeah, i am making myself avoid the simple replies, but i can see how the game as is rewards those. i am wondering how google is going to make that work.

i find watching the dynamics extremely interesting; how quickly one can get a feel for somebody else's strategy.

[akirlu.livejournal.com]

i find watching the dynamics extremely interesting; how quickly one can get a feel for somebody else's strategy.

Yeah, I keep feeling like this is some sort of commentary on the Prisoner's Dilemma and how it really would work if someone is trying to use the game itself as a means to communicate with the other player, which would of course defeat the premises of the game. I think.

[leiacat.livejournal.com]

Looks like the way they are dealing with it is by multiple passes, with the pre-established words being no longer a possibility. I expect that would require people to eventually get to the more specific, but I think I found getting hits on synonyms. (Chart -> map)

The other problem is with some of those images, you have to squint to tell what the heck it is at all, and some of them are really not single-word-describable.

I think it'd be more fun as a game if my brain wasn't primarily on categorization and giving it the most useful rather than the most obvious tags. That is very much a losing strategy.

[pir-anha.livejournal.com]

yeah, the tiny images annoy me a lot. also, i don't care to label menu images that are obviously part of image maps.

the stepwise refinement might actually in the long run lead to useful categorization instead of obvious. i wonder how one can automate the cut-off point though.