over-reliance vs. self-regulation
I found this part the most interesting.
Even though I’m still semi-reluctant about AI, I’ve been getting a lot more benefit by asking general questions instead of asking it to implement things. Prompts like “how can I append X to a Y structure in <language that I’m learning>” instead of “how can write an app that does ”.
I find AI absolutely phenomenal in helping come up with meaningful functions/variable names! It’s also a great tool to discover functions/ libraries that I didn’t know about, which usually leads me to the documentation of the library.
All of this though is in the context of a language that is not too familiar to me. In a language that I’m experienced at ChatGPT feels more like an incompetent coders that I need to be constantly correcting… which, I’d just rather my own code instead of exchanging coding for code reviewing.
Students frequently (n=501, 30%) copied the task description to generate the entire code with no prior manual coding attempts. 😰
I’m not sure I’d consider this negative as the paper does. While it can indeed be a sign of over-reliance it could also be a didactic tool for understanding better the problem. I think many of us might experienced how a problem/task can sound completely incomprehensible and when talking with the author or hearing the instructions rephrase the problem it suddenly clicks and things make more sense. I think that ChatGPT is very good at saying exactly what it’s been prompted but differently, which in some cases can be frustrating, but in some cases (maybe like this one) could be exactly what is needed!
I think the part we’re missing here is how were the kids selected? The linked paper says that they recruited from coding camps, but is this a bunch of kids who wanted to learn to code, or a bunch of kids whose parents want them to learn to code?
Because a kid who doesn’t want to be there taking the shortest path to get through the assignment seems perfectly reasonable. A kid who wants to learn skipping the whole learning part seems significantly worse.
Even though I’m still semi-reluctant about AI, I’ve been getting a lot more benefit by asking general questions instead of asking it to implement things. Prompts like “how can I append X to a Y structure in <language that I’m learning>” instead of “how can write an app that does ”.
I find AI absolutely phenomenal in helping come up with meaningful functions/variable names! It’s also a great tool to discover functions/ libraries that I didn’t know about, which usually leads me to the documentation of the library.
All of this though is in the context of a language that is not too familiar to me. In a language that I’m experienced at ChatGPT feels more like an incompetent coders that I need to be constantly correcting… which, I’d just rather my own code instead of exchanging coding for code reviewing.
I’m not sure I’d consider this negative as the paper does. While it can indeed be a sign of over-reliance it could also be a didactic tool for understanding better the problem. I think many of us might experienced how a problem/task can sound completely incomprehensible and when talking with the author or hearing the instructions rephrase the problem it suddenly clicks and things make more sense. I think that ChatGPT is very good at saying exactly what it’s been prompted but differently, which in some cases can be frustrating, but in some cases (maybe like this one) could be exactly what is needed!
I think the part we’re missing here is how were the kids selected? The linked paper says that they recruited from coding camps, but is this a bunch of kids who wanted to learn to code, or a bunch of kids whose parents want them to learn to code?
Because a kid who doesn’t want to be there taking the shortest path to get through the assignment seems perfectly reasonable. A kid who wants to learn skipping the whole learning part seems significantly worse.
Agreed on your first point. I call this “Using AI as a Reverse Search Engine” and have written about it here.