09/25 - Tic Tac Tuesday

John Kevin [AC1TGWX1U5TKN] Welcome to participate in a short survey (5 minutes) for $0.3 - $0.30 | PANDA

Low

Unrated

Unrated

$7.20 / hour
00:02:30 / completion time

Pros:

Cons:

Qualifications: HIT approval rate (%) > 98; Location EqualTo US;

To read John Kevin's full profile check out TurkerView!​

 

I think the problem is that the questions and answers and what is good seems to be inherently subjective. Although you have five star individuals giving you help with feedback their ideas of what is the correct Q&A is still going to be different than yours.

Also, the last time I got feedback it seemed almost too detailed. Like, I honestly preferred when you would just give feedback on one or two of your hits. But you reviewed like 4 or 5 this last round. I don't know if I think it is overwhelming or an attack? I mean you gave me mostly good feedback, so I am tempted to go with the former rather than the latter.

I think the biggest problem with your hits though is that the ROI isn't worth it unless you do 2, or 3. 2 seems to be the sweet spot. I have a couple of comments about that:

1.) Most mturkers are trying to get good hourly, or they are going to rush through your hits to get them. The sad truth is that, honestly, your hits are filler for a lot of users. So they are putting you on the back burner until they are desperate, and then at that point they are definitely rushing.
2.) Your pictures, honestly, are not very good a lot of the time. (Now don't get me wrong, it has improved drastically with the last batch I did and it's not really feasible to manually review every picture for quality.) If I see a picture that makes it hard to ask questions I will stretch a bit and ask something stupid just so I don't have to return because that effects my hourly.

I think the biggest improvement to get better questions would be if you could have pictures which just detect one person, and say the back of someone's head, and they were removed. It's near impossible not to ask generic questions for these pictures. Even pictures where just two people are looking at each other, and nothing else is going on are hard to ask questions about. What are they doing? Talking. Anything else? No, there is literally nothing else going on in the picture. So now I will ask questions about the room they are in, and why they are in it. That is if the room they are in shows any detail whatsoever. Which some of these are good questions, but a lot of these are stretching really hard. However if you are interested in questions about the room they are in and why they are in it, I would suggest older movies because modern set design leaves a lot to be desired.

Another thing I would suggest would be to remove pictures with motion blur. I don't see how they would improve your AI, from my point of view, and they are equally hard to ask questions for. The karate kid with Jaden Smith is particularly bad about this. I have yet to see a picture from that movie that does not have extreme motion blur. I wouldn't even use that movie, personally, however it may be someone else's favorite movie to ask questions about, so I don't know. Another thing that is entirely subjective.

This is a lot to take in. Perhaps I should have emailed it to you. I don't know.

 

You’ll keep your rating feel free to do as many as you want!

 

btw, @rowan, your hits are actually my favorite to do. Some of this feedback is from the perspective of how other turkers might see it. However, I do rush through some particularly egregious pictures.

 

Matt [ASG8WWXORCBYO] Short survey. Hypothetical manager scenario, evaluate 2 job candidates - $1.00 | PANDA

Fair

Unrated

Unrated

$8.70 / hour
00:06:54 / completion time

Pros:
Fair pay

Cons:
Some brief writing
Has a long exclude list of workers that you have to scroll through to get to the submit button on the HIT page.

Qualifications: HIT approval rate (%) > 95; Location EqualTo US;

To read Matt's full profile check out TurkerView!​

 

CAROLINA

3-0

 

Budget permitting...

Preprocessing HIT with 10 pictures. Rate each 1-10 for how interesting they are. Goes to 2 or 3 turkers (assignments) each.

Images with less than a 4 or 5 average rating get removed from the dataset.

5 cent pay would be acceptable, 10 cent pay would be good.

 

that sounds like rowans I would do

 

This is what I would do. Also, since the picture quality would go up if he did this I would even go further and say that the pay rate for Q&A could probably go down because it would be much easier to do the hits. Go down, but not by much, mind you.

 

4-0

CAROLINA

 

4-1

nashville

 

this is the kind of reddit crap that the ignore button is for

have some respect for the work you do

 

@Bluedays the type of dude to negotiate a pay drop at his annual performance review.

 

Ya'll joking right? Also I'm not sure what the relevance of reddit is.

Also, I think the hardest part of Rowan's is asking questions about shit pictures. The pay rate would still be worth it at .20 each if the picture quality went way up.

 

I also like doing the same amount of work for less pay. In fact, lets pay to do the work. So much more logical than hitting that return button.

 

Hourly would actually still go up, but the budget wouldn't have to increase for the same work. If productivity goes up due to picture quality increase, so does hourly, as long as the pay cut ratio is not drastic.

 

4-4 incoming

 

The only reason I work on mTurk is for fun. Not even money. $2 an hour is better than $0 an hour if you’re having fun