Training AI scams
So about a month ago, as I was surfing Indeed, I found an interesting potential job. Handshake.ai opens up projects periodically and pays people to do piecework at some hourly rate; I was specifically looking at Project Helix, which offered $65/hour basically analyzing AI-generated code and coming up with improvements, etc.
So I signed up and took the CodeSignal test that they prescribed. I made sure to pick my time and place so there wouldn't be any interruptions, and I took a few practice tests to make sure I understood the system.
And I got no response for weeks. After a while, I saw that there was a status change on the Handshake dashboard, so it seems that the first test I took wasn't recognized or accepted. Weird. I gave up and moved on.
Within the last 2 days, I was contacted by Handshake.ai, telling me that Project Helix was "finally open" (you mean it wasn't before?) and they would dearly like me to take the CodeSignal assessment. They said I looked like a strong candidate, maybe from parsing my résumé.
So this time, I said "screw it" and took the damn test. Again. Some of the questions were the same, but the final coding question was different. And because I didn't pace myself this time, I didn't have a whole lot of time to refine my final answer. I was failing some of the hidden test cases, but I was passing the ones I could see.
Also of note, that final test question was the only actual coding problem on the entire test; just about everything else was some kind of free-form essay. This is an important detail.
Within about 5 minutes of submission, I got an e-mail with a rejection. Said I didn't pass the assessment and would not move forward.
So someone please explain to me how they were able to grade me so fast the second time around? I doubt that a human being looked at it. Either that last question counted for more than I realized, or they had a machine looking at my essay answers and "grading" them. I'm pretty sure my responses didn't completely suck.
The other thing is, the final question was a Java programming exercise, and in the instructions they tell you to take some value and compute the modulus with a large-ish value (10^9 + 7) and return that result. Well, everyone knows in Java that you compute the integer modulus using the % operator. Except that the version of modulus that the test writers expected isn't that, because the modulus operation they expected doesn't handle negative values the same as % does. This caused me a lot of grief. The specification (instructions) didn't say which kind of modulus to take, so I had to reverse engineer it from the test cases that I was actually able to see.
My suspicion is that whoever wrote this test was writing it from the standpoint of another programming language (probably Python), and just expected me to know that mathematical function instead of the way modulus is implemented in just about every C-like language out there. Yes, there's a formula for computing the mathematical modulus for negative values in languages like Java that return a negative modulus result: ((a % b) + b) % b
I certainly couldn't remember the precise formula (had to look it up just now and do the obligatory face palm – couldn't do that during the test). My point is, this is pretty misleading for a Java-language problem.
I didn't run into this problem with the last CodeSignal test, and even HackerRank doesn't pull this level of "gotcha" crap.
Normally, I don't go in for piecework – it's kind of skeezy. But it sure would have been nice to have some money coming in while looking for a real job.




















