[varunnrao]

I always find that whenever such topics come up for discussion Ted Gray had the best take - https://theodoregray.com/BrainRot/.

With AI and specifically LLMs, I feel many of the points raised here about 25 years ago still stands. Any LLM based learning isn't meant to push students into thinking about the problem or even guiding students along the path to the correct answer. It's mostly about returning some answer -- mostly right but not really of a good quality. As pointed out in Ted Gray's article, any assisted learning mechanism should work towards making the student think about the problem at hand and reason their way through it. It's an aide to the human teacher because, it being a machine and most importantly software, can adapt to each and every single student's ability and work at the student's pace instead of 3 periods of an hour each per week. An LLM by definition is the opposite. It gives you the next most probable character without any idea of what it's saying. There is a lot of work being done on this ofc, but what's available today isn't a reasoning aide/teacher.

[throwaway89201]

Thanks a lot for posting this. I agree that many of the points still hold, and as a (small business) author of educational software, reinforce my ideas about what that software should look like, especially concerning open-endedness and "capable of doing senseless things if asked".

I do think that the arguments about violent games either seem a bit dated or simply don't hold at all beyond operant conditioning. Especially in Europe, children or adults are hardly in the position that such 'reptilian brain' conditioning matters.

[Balgair]

I'm struck by how the description of the more ideal case is just, well, mostly montessori style education. But a bit cheaper.

One wonders by how much we'd have to increase the budgets of the various school systems to have mostly montessori style schools and not have to even worry if we'd like better schools.