Separately, I wonder whether you'd find something like this more trustworthy: https://kidsdiscover.com/shop/issues/roman-empire-for-kids/

And, if so, what signals do you think makes it more trustworthy in your eyes?

Great question. Thankfully most of the topics I'm covering here are well covered by trusted sources that are freely available on the internet. This is particularly true about dates (e.g. when a particular philosopher died). Separately, many of the articles are less about the specific history of each topic, but about how the topics fit together and what you as a reader might like to think about as a result. These parts aren't stating facts (which can be true or untrue) but explaining that there's a different way to look at something (which is matter of opinion or perspective).

I figured other people might find those AI-generated articles helpful, so I put them here: https://philosophy.ocaho.com/

There's a search box at the top.

Comments URL: https://news.ycombinator.com/item?id=48530657

Points: 11

# Comments: 5

That means two trips to the library. And it means you can't use the book that same day. This is fine for fiction, but not if you want a book because you're going to study it to aid your learning or your work.

If it's fine for a book to arrive in a few days why have a library with publicly browsable shelves at all?

Am I meant to then override that by increasing the center channel volume so it's louder than the other speakers?

Or raise the system volume?

Pearson cannot force a student buy anything. But you can.

The moment you decide your course will incorporate MyLab, McGraw-Hill Connect, Cengage MindTap, or WebAssign, the student is on the hook to buy the access code, by buying a brand-new copy of the book. The access code isn't a freebie that comes with the book. It's the main reason to buy the book.

That access code does one thing: it moves the grading work. Grading used to be your job or a TA's, and it was paid for out of the tuition fees the student already paid to your employer.

Now you're double dipping: you make the student pay tuition fees to attend your classes, and then you make the student pay your outsourced auto-grading provider.

I highly recommend reading the book, even if you expect to disagree with every fact, statistic, opinion and conclusion.

Of course, the above paragraph isn't perfectly accurate. It's based on my impression of a book I read a couple of years ago, polluted with my own biases, other things I've read on the topic etc.

But this part misses the point:

"As someone who makes use of AI quite a bit in my own learning, I can say that it’s still relatively weak at having a good model of an individual’s skill gaps and conceptual weaknesses."

It seems like he is expecting a chat-based LLM to maintain a model of the user's skill tree. But it wo:

- create a detailed skill tree for whatever subject

- have the user try to apply the skills

- store the user's mastery level for each node, in some structured format

This isn't something ChatGPT is going to do if you just starting chatting with it.

But you can design a system to do it, which is what the Math Academy folks have done.

Edtech tools don't have to have user-facing AI. They can use AI under the hood, or use no AI at run time at all.

- "Responsible for the code", i.e. understands it, can explain it, can modify it on request, is the right bar for a working engineer using an LLM to go faster. But it's the wrong bar for a student, because in a course the deliverable isn't the point, the practice is.

I presume that as course like 'Parallel Computing' exists to build the habits of reasoning about decomposition, contention, race conditions, strategy tradeoffs. Those habits come from doing the reps, the same way you don't learn to write by reading good prose. Your two "successful" groups outsourced exactly the reps the course is (or should be) there to build.

Being able to follow and tweak a solution proves comprehension, but comprehension is a much lower bar than producing the solution.

Your students may understand this particular solution well enough to explain it. But did they learn enough that they could tackle a different problem with a different shape, without using an LLM?

That's why I raised the low N.

Separately, there's a clear downward trend in the number of grades being awarded in the fall semester. Perhaps students have the option to take the course pass/fail?

In the most recent CS10 cohort (the one in which 35% of grades were an F) only 34 students were graded.

If you're going to look at intro classes, why not look at the Fall semesters, which have much higher enrollment?

             # grades   % F grades
 Fall 2019        268        0.37%
 Fall 2020        259        8.49%
 Fall 2021        342        3.22%
 Fall 2022        218       13.30%
 Fall 2023        194        7.22%
 Fall 2024        169        1.18%
 Fall 2025        146        2.74%