[matsemann]

While I get this stance (the books are often written with students in mind), it's sometimes a bit hard when reading this kind of material while not enrolled. I don't have any TAs, professors, peers etc to ask or grade my stuff when working my way through a text book in my spare time. How can I be sure I understood stuff correctly?

For me it's ok, I guess. I have graduated years ago and it's strictly for fun. But I wonder if it also creates a kind of artificial divide for those not able to attend a university but could have gotten good use of this kind of material by self study.

Edit, found this quote from him on an earlier discussion which summarizes it nicely:

> I'm honestly seriously torn about this. There is a serious tension between pedagogical needs of students in formal classrooms and the pedagogical needs of self-learners. I've chosen to aim for the former. Yes, I know it's a bummer.

[bhrgunatha]

That's what was good about the early stages of sites like Coursera, Udacity and Edx - they had quizzes, exercises and exams that are auto-graded that provided that kind of valuable feedback for all tiers.

I think they've all evolved now to where that is only available for paid tiers.

One innovation I really enjoyed was marking and commenting another student's answers, once you'd submitted your own. There are flaws but it's a valuable variation on graded feedback.

[stevegalla]

> I don't have any TAs, professors, peers etc to ask or grade my stuff when working my way through a text book in my spare time. How can I be sure I understood stuff correctly?

I definitely agree with you on this. However, there are three things self-learners (or even students) can do.

1. Find a textbook with worked solutions or a different presentation of material.

2. Look for course material that has problem sets with solutions, worked tutorials, or worked examples.

3. Learn numerical or other methods to evaluate your solutions. I know this isn’t possible for proofs, but if it’s an algorithm’s run time, I don’t see how implementing the algorithm and plotting run times over different input sizes wouldn’t be a good way to see if you have the correct run time.