[lolinder]

I took a deep learning course in late 2019, during which we implemented transformers as described in Attention is All You Need and fine-tuned GPT-2. The output was amusing, but useless, but I still remember the basic principles.

Now, a few years later, transformers are the tech, and GPT-2's successors are the most hyped technologies of the century so far.

All of which is to say that I wouldn't assume that coursework without immediate application is useless. I'm in a much better position to jump in on the latest AI stuff than I would be if I hadn't taken that course.

[imranq]

I mean if someone likes the subject, then they should invent a project that forces them to build something or write something up. Otherwise its easy to "fake" learning by doing the motions on a bunch of tutorials and quizzes.

[caddemon]

It really depends how good the course is, with well-designed problem sets/project prompts you can't really fake learning (assuming you actually complete them). Is it going to be totally exhaustive of everything you may need to know in practice? Obviously not, but no single project will be either, especially not for such a broad field as machine learning.

Independent projects can definitely be a great way to learn, and yes many courses are shitty. But it is also very possible to take a good course and walk away with new knowledge you didn't even realize you needed. Some of my favorite projects actually started with an idea from a course, and then I learned even more in order to further expand on it. Synergy between project-driven and course-driven education can be a powerful iterative process.

[victor106]

>I took a deep learning course in late 2019, during which we implemented transformers as described in Attention is All You Need and fine-tuned GPT-2.

Which course was this?