Comment by hintymad | Hacker News Reader

Comment by hintymad | Hacker News Reader

I used to take a lot of courses on cousera and EdX. I still take some here and there, but not as many as before. Some of the courses are amazing and unbelievably rewarding, like Daphne Koller's Probabilistic Graphical Models, Robert Sedgewick's Analytical Combinatorics, and Gerald Sussman's course on system optimizations. I'm very grateful for such learning opportunities. Unfortunately over time, I also found that these courses had diminishing returns for the following reasons:

- Due to the nature of MOOC, the assignments are largely either multiple-choice questions or programming assignments that merely asked students to fill in some blanks in some functions (there are a few exceptions of course). What a descent US university does really well is challenging students with tough yet insightful and inspiring assignments. That's how students learn deeply and retain the knowledge, at least for me. Merely listening to lectures and ticking off a few ABCDs hardly helps real learning.

- Lack of feedback. A university course assigns TAs, gives tutorials and office hours, grades assignments with detailed feedbacks, and it is so much easier to form study groups and have high bandwidth discussions. MOOCs try their best to offer such help, but they don't work as well or at least not as conveniently.

- Many courses are watered down. For instance, Andrew Ng's ML course on Coursera is far less rigorous than that (229? I forgot) offered in Stanford? The course is great for students to gain some intuition, but I'm not sure if it's good enough for one to build solid ML foundations.

I completely agree that trying to get an education/certification model with no feedback and simple, robotic questions is completely useless.

What's even worse is that almost none of the MOOCs used the strengths from online classes. There are other ways to learn than a hour-long video of a person talking to a webcam.

It's funny how a big part of Andrew Ng's classes is waiting for him to write text with mouse as if he were using the world's worst whiteboard; he could have prepared properly-drawn figures in advance.

students are more engaged with the lesson when the teacher handwrites compared to when teacher uses slides / ready-made material

They could at least use a Wacom.

>They could at least use a Wacom.

This, when they draw with a mouse, even after tons of lectures, it always looks like shit. The good ones use tablets and change colors and thicknesses and such as they go.

To be honest, instead of wasting time on writing, the professor could share anecdotes from history, his life, industry, and other aspects i.e. the social aspect of doing science and research. I really find those interesting than mere cut and dry exposition of concepts.

Worthless is greatly exaggerating imho. I learned quite a bit from these courses even though they were far from optimal.

Many studies have shown watching things happen over time is much more useful for the human mind than being presented it completed.

That being said, one way to achieve this is to play things backward or occlude detail while you get to the final creation.

> It's funny how a big part of Andrew Ng's classes is waiting for him to write text with mouse as if he were using the world's worst whiteboard; he could have prepared properly-drawn figures in advance.

My personal experience actually showed otherwise. It was more effective for me to learn if instructors write on a whiteboard to gradually develop what they teach. I guess that's because when an instructors writes a whiteboards, students will know exactly what she focuses on all the time, and the writing speed matches the speed of understanding. In contrast, a professor in my university was a big shot on operating systems. He used well prepared slides and he talked fluently, yet I got lost in almost every class.

Udacity had a great Python course back in the day (Programming 101, for Python 2.7)

It had vidoes and then a REPL would drop down and you would continue writing your program, test it against test cases and then a new video would go on for a few minutes explaining the theory for the next step.

I can see why not many try and do this. It obviously took a lot of work, both technically and pedagogically to set up the course and problems.

I remember taking Andrew Ng’s (delightful) Coursera ML and believing I knew ML.

Then I took a Columbia ML graduate course IRL: it was like being hit by a train.

How would you rate the prerequisites of each course and level of material covered? I've never taken Andrew Ng's ML class, but the impression I get online is that it's great but it's always hard to tell from these positive reviews if the course is just an introductory exploration or something more in-depth.

229 is an intro class for students with no AI or CS theory experience, and basic multivariable calculus and linear algebra.

229 is a 2xx class for advanced undergraduates and beginning graduate students. It's a first course in machine learning with no AI prereqs.

What grad class did you take?

https://www.cs.columbia.edu/education/ms/machinelearning/

https://www.cs.columbia.edu/~jebara/4771/handouts.html

In my experience, there’s a vast difference between “education” aimed at the individual and what is delivered in accredited academic courses. The commercial aspect / tailoring to get people to buy and stick with it / no doubt is a factor.

To me that is in large part explained by the fact that most MOOCs are introductory, unlike most graduate courses.

Coursera got enshittified like crazy. The first couple of years had legit college courses on it. Then it became all about micro degrees and courses with twenty minute lectures.

There's still some really good stuff out there. I just finished the General Chemistry courses on EdX and they're really good (with a very quiet discussion forum that's still visited by MIT staff). The Finance MicroMasters was also excellent and had active TAs on most courses. Exercises on all these courses were generally very high quality.

Another great online course that I recently took is this one on parallel computing [1]. It's not on Coursera/EdX but uses a custom platform, and I would say it goes beyond "fill in the blanks", the assignments are really challenging and have a lot of depth.

Compared to 5-10 years ago the trend is unfortunately definitely downwards though. A lot of great courses are archived and far fewer are being added than there were in the past.

[1] https://ppc.cs.aalto.fi/

Do you remember the name of Sussman's course on system optimizations? Can't seem to find it online

I forgot the name. It may be called system engineering or something like that. The focus of the course was on parallelization. The instructors spent great deal time on work stealing queues and parallel divide and conquer.

Feels like LLMs could be used to scalably grade more complicated assignments than multiple choice tests.

Reply on news.ycombinator.com