Something I think you neglect in this comment is that yes, you’re using LLMs in a responsible way. However, this doesn’t translate well to school. The objective of homework isn’t just to reproduce the correct answer. It isn’t even to reproduce the steps to the correct answer. It’s for you to learn the steps to the correct answer (and possibly the correct answer itself), and the reproduction of those steps is a “proof” to your teacher/professor that you put in the effort to do so. This way you have the foundation to learn other things as they come up in life.
For instance, if I’m in a class learning to read latitude and longitude, the teacher can give me an assignment to find 64° 8′ 55.03″ N, 21° 56′ 8.99″ W on the map and write where it is. If I want, I can just copy-paste that into OpenStreetMap right now and see what horrors await, but to actually learn, I need to manually track down where that is on the map. Because I learned to use latitude and longitude as a kid, I can verify what the computer is telling me, and I can imagine in my head roughly where that coordinate is without a map in front of me.
Learning without cheating lets you develop a good understanding of what you: 1) need to memorize, 2) don’t need to memorize because you can reproduce it from other things you know, and 3) should just rely on an outside reference work for whenever you need it.
There’s nuance to this, of course. Say, for example, that you cheat to find an answer because you just don’t understand the problem, but afterward, you set aside the time to figure out how that answer came about so you can reproduce it yourself. That’s still, in my opinion, a robust way to learn. But that kind of learning also requires very strict discipline.
So, I’d point back to my comment and say that the problem really lies with how it’s being used. For example, everyone’s been in a position where the professor or textbook doesn’t seem to do a good job explaining a concept. Sometimes, an LLM can be helpful in rephrasing or breaking down concepts; a good example is that I’ve used ChatGPT to explain the very low level how of how greenhouse gasses trap heat and raise global mean temperatures to climate skeptics I know without just dumping academic studies in their lap.
Your example at the end is pretty much the only way I use it to learn. Even then, it’s not the best at getting the right answer. The best thing you can do is ask it how to handle a problem you know the answer to, then learn the process of getting to that answer. Finally, you can try a different problem and see if your answer matches with the LLM. Ideally, you can verify the LLM’s answer.
Something I think you neglect in this comment is that yes, you’re using LLMs in a responsible way. However, this doesn’t translate well to school. The objective of homework isn’t just to reproduce the correct answer. It isn’t even to reproduce the steps to the correct answer. It’s for you to learn the steps to the correct answer (and possibly the correct answer itself), and the reproduction of those steps is a “proof” to your teacher/professor that you put in the effort to do so. This way you have the foundation to learn other things as they come up in life.
For instance, if I’m in a class learning to read latitude and longitude, the teacher can give me an assignment to find
64° 8′ 55.03″ N, 21° 56′ 8.99″ Won the map and write where it is. If I want, I can just copy-paste that into OpenStreetMap right now and see what horrors await, but to actually learn, I need to manually track down where that is on the map. Because I learned to use latitude and longitude as a kid, I can verify what the computer is telling me, and I can imagine in my head roughly where that coordinate is without a map in front of me.Learning without cheating lets you develop a good understanding of what you: 1) need to memorize, 2) don’t need to memorize because you can reproduce it from other things you know, and 3) should just rely on an outside reference work for whenever you need it.
There’s nuance to this, of course. Say, for example, that you cheat to find an answer because you just don’t understand the problem, but afterward, you set aside the time to figure out how that answer came about so you can reproduce it yourself. That’s still, in my opinion, a robust way to learn. But that kind of learning also requires very strict discipline.
So, I’d point back to my comment and say that the problem really lies with how it’s being used. For example, everyone’s been in a position where the professor or textbook doesn’t seem to do a good job explaining a concept. Sometimes, an LLM can be helpful in rephrasing or breaking down concepts; a good example is that I’ve used ChatGPT to explain the very low level how of how greenhouse gasses trap heat and raise global mean temperatures to climate skeptics I know without just dumping academic studies in their lap.
Your example at the end is pretty much the only way I use it to learn. Even then, it’s not the best at getting the right answer. The best thing you can do is ask it how to handle a problem you know the answer to, then learn the process of getting to that answer. Finally, you can try a different problem and see if your answer matches with the LLM. Ideally, you can verify the LLM’s answer.