Comment by nothis | Hacker News Reader

Comment by nothis | Hacker News Reader

Parent Full thread

Something I noticed is that exams are what we're always coming back to in terms of education. Employers want a confirmation that you learned something. That's a university degree. Even with respected websites offering confirmations, how can you trust that and how do you keep track of every website's quality?

I noticed this about all those "machine learning courses" that are getting posted here. Facebook and Google basically hand them out for free, hundreds of hours worth of quality learning material. Top universities let you watch all their lectures online. So what's the catch? Why aren't we all machine learning experts already? Because it's hard to muster the energy unless you commit to earning a degree by, yes, studying for exams. Turns out that's actually thousands of hours of work so the "convenience" part of online courses kinda starts to fade.

I definitely noticed that the quality difference between a good university and a bad one is often about how much they value "education" beyond just being an "exam machine". Presenting facts from the point of view of someone new to them, presenting methods of how to deal with edge cases and traps, taking one-on-one tutoring sessions seriously.

But yea, the end result is getting confirmation for having passed a bunch of exams.

In addition to the scarcity of what I'd call "deep dive" materials, companies like Facebook and Google are frequently trying to encourage platform adoption.

MOOCs for purely pedagogical reasons are sometimes hard to find in ML, not only for the aforementioned reasons, but creating a MOOC is simply resource intensive. It follows that many of the producers that truly value education are actual universities. Hybrid models like Coursera and Udacity are still trying to figure out how to get out of the red.

All of this to say that I'm still very grateful for content creators who go beyond simple framework tutorials.

I don't think this is an accurate assessment of education in the ML space. There are tonnes of people who have self-taught themselves the material. The issue I see is that employers do not trust non-university credentials, and so there is a strong hiring bias towards PhD graduates, regardless of whether it is warranted.

I think a big part of this comes from the lack of emphasis on group projects and technical communication in MOOCs. How many MOOCs have you seen where you're actually encouraged to work in groups, and then eventually present your work?

Even though the concept of an MOOC has been around for quite some time now, it's only recently that their curricula at the course level is actually consistently on par with what would be seen in a good university environment (think like "Top 50" University).

I had a whole post in startupschool.com about this. I was unable to find anybody agreeing with me. Glad you made me feel less alone in this world. And TBH from all the excitement of being in SuS, I think one might launch a startup with this idea

Most self-taught people I have met have come up with their own projects, or at least played with Kaggle, as people are usually advised to do. MOOCs are only a small part of the options people can and do exercise.

Self driven education has always existed. At least as long as you could check out books in the library. The Internet made it more convenient, but it didn't make it any easier. Most people thrive in a structured environment with superiors and peers. That's why universities are so much more successful than libraries.

There's also the problem of knowing what you don't know. If an expert in the field is educating you they will give you a full picture of the field and what's important in it. But if you're self taught you can miss entire areas and never realize their importance from reading academic sources.

I'm not against self teaching or MOOCs at all. I just think they need more structure and require more commitment to be as useful as university is to society. They're a bit too casual, requiring much more self drive which is harder than just setting deadlines and exams.

One thing I’ve realized is that there is a lack of intermediate or more difficult material avaiable.Lots of these courses work on toy examples with data already cleaned up and organized.

I wish there were more courses breaking down the algorithms in the latest research papers and actually implementing the models .

Reply on news.ycombinator.com