Life changes, or how the course went from being too high level to too low level from my perspective

I remember taking the fastai course for the first time

“OMG this is so high level, how is this even learning? Where are the derivatives?”

That was at a time when I studied machine learning.

I didn’t compete at Kaggle. I didn’t have an ML job. I watched lectures, read books and papers, pondered set theory and the construction of real numbers, I didn’t even realize that PyTorch (or anything else with autograd) existed and I wrote code like this.

Fast forward 5 years to this week’s lecture. Having worked in ML for the last 4 years or so

“OMG this is so low level. Ahhh so this is how we calculate the bias term, and this is where we stick with torch.no_grad(). Ahhh, I see.”

I found this experience quite surprising :slight_smile:

What is the meaning of all this?

On one hand, it might be that what you do in the workplace, how you bring value to the world, is very far removed from the perception of learning instilled in us at school.

The acquisition of skills that make us valuable in the real world might not feel like learning at all!

Case in point – the job that I started just recently. I am getting so much value in my work from

  • being able to identify what is important to focus on (we practice this as we learn a new way of reading papers which will most likely be discussed later in the course)
  • how to learn from jupyter notebooks created by others (this literally mirrors the process of doing the fastai homework!)
  • being able to set up my own work environment, install software libraries, navigate to parts of the codebase that are relevant without giving it too much thought, all of this has become second nature to me and is something that we practice in this course

And the list above doesn’t even mention probably the biggest component of what I do on a regular basis, that probably led to me having ML jobs in the first place, and that is technical writing! Something that felt like play, that I only continued to practice because of fastai, but that actually turns out to be the real work despite seeming to the contrary! (a great blog post on how to get started)

Another lesson I derive from this experience is that your perception of what is covered in the course speaks more about you and where you are in life than about the course itself.

The best thing one can do for oneself is to approach the course with an open mind. Follow the instructions and withhold your disbelief for as long as you can.

From that it springs, that the best way to evaluate this course is not by how you feel about it now but by the effects it produces for the ones who have taken the course some time ago.

And the effects are quite good :slight_smile:

I wanted to share this as a tweet but I got second thoughts after the fact. Yes, the anxiety about sharing things online might never go away!

Very excited and grateful to be on this journey with all of you :slight_smile: :pray:

23 Likes

Great post, Radek! Thank you!

1 Like

Wow. I missed that when you posted it. Got a link?

1 Like

I got cold feet so I deleted it. :slight_smile: It is a bit tough for me to say if a tweet I write is appropriate or not, if I am considerate to the other people that I mention, among other things.

Also, the best way to use twitter is when you assign very little weigh to your tweets, when they are dispensable in a way, where you don’t worry too much about any single thing that you write. I am obviously not there in my twitter game :slight_smile:

Anyhow, seems that tweet might be innocuous or with some luck even useful to someone :slight_smile:

Queued it up for today at 1PM (works nicely with our friends in US, where I can automatically retweet it in 12hrs for their perusal :grin: untested theory that this does work, but makes me feel very pro :wink: )

Will post a link to it here once it goes live

As a side note, I would like to add that the fastai community, the fastai forums, make for a really great place to write :slight_smile: With the broader Internet, it can get scary at times, but we are so welcoming and understanding here, there really is no downside to sharing your thinking here. That is a wonderful feeling, makes playing around with new things and ideas really easy :slight_smile: I quite miss that vs writing in the open, now that my audience is slightly bigger.

4 Likes

Thanks for sharing your insight here Radek. Reminds me a bit of Paul Graham’s “Beating the Averages” essay discussing the Blub Paradox of programming languages.

The corallary is that the only languages worth learning are those that change the way you think about progamming, shifting your perspective to see more of the Blub landscape. So a plug… the biggest impact in this regard for me was Smalltalk inspired Pharo. Its worth doing a hobby project in.

From my current naive Notebook newbie perspective, one thing missing from online Notebooks is not being able to browse contexts up a debug call stack (perhaps others can clue me in.) PharoLambda shows what is possible. After completion of this course I’m going to try porting fast.ai to Pharo.

4 Likes

Thank you for sharing such a thoughtful post. I sometimes wonder if I’d ever be able to be a tenth as competent and as good at things as some of the fellow travellers in this community but I do agree that coming back to this course and getting something from it does really depend a lot on where you’re at in your journey and it’s always something new.

I don’t use ML/DL in my daily work, and I’m not sure if I’ll ever do so, but the journey is definitely worth it.

3 Likes

And we are off to the races! :slight_smile: The tweet is now ingested by the beast that Twitter is :grin: Here it is.

1 Like

Glad you brought it back and thankful to be a part of this community where I get to meet and work alongside folks of your caliber and character.

3 Likes