Comment by mattlondon
2 years ago
Well the point is to make them think about the prompt. I don't care if someone prompts to come up with a second prompt to get the answer, they need to input and refine to make it fit the goal. So by setting fairly stringent objectives for what the AI needs to generate, you need to be quite specific to engineer the prompt and think about what it needs. And then you need to look at what it said and qualify it somehow to decide if it is good enough - sure you could use a LLM for that too, but what is the prompt for that? You'll need to teach the second LLM what to analyse in the output of the first LLM to meet the requirements of the assignment, so the prompt(s) becomes the essays of yesterday in terms of academic work/thought for the student.
That's the point of what I was getting at - they're going to use AI regardless, so at least work with it rather than against it.
Right. But what I mean is if the final evaluation will require an in-person discussion, then couldn't we skip the text generation step, and just have the discussion? Homework as a tool to make students think needs to be drastically different, since students will always find loopholes to avoid it, and we can't evaluate it in isolation anymore (i.e. a teacher can no longer grade it on its own).
Doing live discussions obviously requires much more time and effort, but it is also much more engaging than doing homework, so the chance of actually learning something is higher.
There's also an argument to be made that homework in general isn't conducive to learning, and that education should take place primarily in schools. When you consider how much future generations will rely on AI in their everyday lives, this argument becomes even more compelling.
So I think this is our chance to drastically change how we think about education, rather than try to shoehorn a groundbreaking technology into our outdated methodologies.