4 ms·
The point that you're (and everyone is) glossing over here is relative positions on the skill gradient. A first grader probably would be prohibited from using
by treyd 1y ago
The point that you're (and everyone is) glossing over here is relative positions on the skill gradient.
A first grader probably would be prohibited from using any kind of calculator on arithmetic tests, 4-fn or not. But 8th graders are usually permitted scientific (non-programmable) calculators.
As you go up in grade level, you "get access to" calculators capable of functionality at the level below you. Because the point is that when we're educating students we want them to actually learn the subject matter, but once we've deemed them to have understood it and we have them move onto the next goal, we give them the tools to make that prior goal easier. We lessen the burden of the little mechanical concepts they already know so that they have an easier time becoming familiar with the next more advanced concepts.
AI systems are so much more advanced than what's capable on a TI programmable calculator. It's hard to draw clean boundaries around the tiers and enforce them by telling the model "help the user with tasks of tier 1-4 but not 5+". That's the issue, that it's really infeasible to strictly use them strictly as learning tools. You can almost do it with a lot of self-discipline and self-reflection to analyze your own workflow, but it's not generalizable across domains.
- pishpash 1y agoWhy can't they be restricted to produce only the concepts of a grade below you? It sounds doable and is actually a great idea.
- tomsmeding 1y agoHow are you going to ensure that it is impossible for the student to work around whatever measures you take?
- estearum 1y agoCan LLMs be reliably restricted to produce any specific subset of content? AFAICT they're still consistently jailbroken.
- wat10000 1y agoIt’s easy to do at the syntactic level by controlling the sampling. For example, it’s easy and common to restrict output to be valid JSON but just not allowing any tokens that would make it not valid JSON. But reliably restricting output at the semantic level is very much an open problem.
- brudgers 1y agoI mentioned the continued ban on programmable calculators in many academic contexts. Those contexts still include some portions of undergraduate education. This is fifty years after the introduction of programmable calculators. Realistically, the answers the students gave the teacher were probably motivated by the practical benefits that come with giving teachers the answers they want to hear…bullshit questions are likely to produce bullshit answers. It’s not like first year college students haven’t had twelve years of academic standards moralizing talked at them.
- the_af 1y ago> Realistically, the answers the students gave the teacher were probably motivated by the practical benefits that come with giving teachers the answers they want to hear…bullshit questions are likely to produce bullshit answers. Like I said to you in another comment, this is a fundamental misunderstanding of the article. At least one of the polls was anonymous, too. There were no punishments in the class for using AI, in fact at least a couple of the students later revealed they had been using AI for everything. This wasn't a "bullshit inducing poll", it was an experiment in perception vs reality, and how they modified those perceptions after the experiment had run its course.