11/13 - Wide Wednesday!

I want to wish happy holiday to all of fbturkgrind

 

Samantha Joel [A3MMG46JS5OPJP] Answer a short survey where you rate selected photographs of faces on one provided trait - $0.50 | PANDA

Good

Unrated

Unrated

$10.53 / hour
00:02:51 / completion time

​

Pros:

Cons:

Qualifications: HIT approval rate (%) > 95; Location In CA, US;

To read Samantha Joel's full profile check out TurkerView!​

 

But are you still a doctor?

 

Anyone know how a requester can see how workers answered a qual test? Like if they want to see if any questions were commonly missed to refine the instructions, or give specific worker feedback?

 

On MTurk's qual test template or a custom qual HIT?

 

Template, with auto grading if that matter

 

Sticky [A1I5WZRIXY5GS9] Fast and easy content viewing (few minutes for possible $0.60) - EEwQRr0IAKv4dFaxC8Pzp - $0.40 $0.20 bonus | PANDA

Generous

Unrated

Unrated

$14.30 / hour
00:02:31 / completion time

​

Pros:

Cons:

Qualifications: Age <= 70; Location EqualTo US; Age >= 18;

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

 

how much do you guys tip the instacart driver (or other grocery delivery driver)?

 

Title: Evaluating Faces(~ 10 minutes) | PANDA
Requester: Stanford GSB Behavioral Lab [A3OSXTUM1QEXNY]
TurkerView: [ $14.08 / hour ]
Description: Earn $1.20 in 10 minutes!
Duration: 50 Min
Available: 61
Reward: $1.20
Qualifications:

• Total approved HITs GreaterThanOrEqualTo 100,
• Exc: [5829544-181840] DoesNotExist
• HIT approval rate (%) GreaterThanOrEqualTo 95,
• Location In US,

[3ERG9KP7KA68VNDTZNYB8ITSQHLHCM]​

 

Ew, honestly I'm not sure @Tjololo ?

 

Can I get some suggestions on where I should order food from today?

ty

 

Same place I'm ordering from, bottom of the deep freezer.

 

Does it take debit cards?

On the real I want my annual Chipotle burrito but the nearest Chipotle is forever away

 

you are catching on

 

im down

 

Which hit is that?

 

I suppose it's good that some requesters are actually getting hip to the idea of checking to see if respondents are taking the time to provide plausibly human responses, and it's definitely good that Mturk is proactively and publicly investigating reports of obviously inappropriate use.

I guess I don't understand why, especially for students posting public academic surveys, every study isn't carefully investigated, if only as part of the training of the student by their academic adviser to teach the lesson that a percentage of people are always going to attempt to defraud any system.

The real risk to surveys and opinion polls, although obviously not just limited to rinky-dink mturk academic surveys, is somebody taking the time to set up a generative text bot with an adversarial network to scrub through a corpus of general questions and spit out a few thousand plausible responses for any given prompt, retrain or just pass it through a few human editors, then package it up and sell it to unscrupulous workers (or, use it as a coordinated tool to alter research, set up plausible legends for fraud, etc.).

Even GPT-2 1.5B is probably already adequate to slip past 90% of audiences in 90% of scenarios with minimal human oversight; just wait until somebody malicious decides to spend a few weeks on the problem.

 

I could eat taco bell