Q1 of 2024 in review: How I used the “12 week year” for studying ML

Updated

I’m the first to admit that most productivity techniques don’t work for me. I read about them, watch videos on YouTube, and sometimes I even start applying them (looking at you, Getting Things Done, with your dozen lists), but they never properly integrate into my daily life.

In January this year, I gave productivity advice another try with the “12 week year”. This technique was essentially invented for procrastinators like me. In this blog post, I will share the goals I set for myself for the first quarter and how much progress I made – with a significant focus on my self-teaching goals around LLMs and NLP in the field of machine learning.

Table of Contents

What is the “12 week year”? A short introduction

At the center of this productivity framework is an observation most of us have probably made in life: at the end of the year, we all start scrambling and frantically trying to reach our goals:

  • At the end of the fiscal year, businesses suddenly do everything to increase their revenue.
  • At the end of the semester, we suddenly start studying.

Often after this sprint, we reevaluate our goals for the next year. What if you could shorten this feedback cycle? Achieve goals 4x per year instead of 1x? Adjust your goals every 3 months to ensure you are really working along your ideal path? A dream scenario for serial procrastinators and people, like me, that are easily overwhelmed by goals that are too big and vague.

Created by Brian P. Moran and Michael Lennington, this idea is published in a book called “The 12 Week Year” and you can also find more information on the accompanying website:~https://12weekyear.com/~

After each week, you recap: How many of my daily and weekly actions did I achieve? Increasing the number of completed actions is the primary focus. It’s not about reaching your goals, but improving your commitment to your plans. You also track some metrics to see if your actions are having the desired effect, to avoid persisting in the wrong approach for too long (e.g., track your 1km time to check if going for a run 2x per week is actually helping).

My study goals around LLMs (Yes, 3 goals. Yes, that was a lot.)

The authors of the book suggest picking three goals, one for each area: Health, Personal, and Professional. I loosely oriented myself along those categories, completely disregarded the rule of three though, and picked six instead. As you can imagine, that didn’t work out too well… So, the following is from the “Personal” area and should have been just one goal:

1. Reading “Build a Large Language Model (From Scratch)” by Sebastian Raschka: My primary goal was to enhance my understanding of how LLMs are built and how they function. This book, currently being published in Early Access chapter-by-chapter, focuses precisely on that by breaking down all the components of an LLM and coding a simplified version. It starts with tokens and attention, going all the way up to training a generative model. My target was to thoroughly read the first three published chapters.

Weekly actions: Read daily (started with 30min/day, revised that to 15min/day)
Results: Finished the three chapters. Actually, only read on 31 out of 84 days.
Mistakes I made: I didn’t schedule the reading and I stressed myself by setting daily reading. I always did it very late at night after procrastinating. As we can see, reading daily was not necessary whatsoever.

2. Read 8 scientific papers: I anticipated needing less time for the book, so I added the paper reading goal. The idea was to gain an understanding of current challenges in the field of understanding LLMs, as well as best practices for RAG, evaluation, and security.

Weekly actions: Read daily (started with 30min/day, revised that to 15min/day) (see above)
Results: Finished 3 papers. Started a bunch of others. Actually, only read on 31/84 days.
Mistakes I made: See above for general reading. However, I also should have selected the papers more carefully and at the start of the quarter. A lot of papers that I started reading turned out to be not that relevant or simply not at my current level of research understanding, which really hurt my motivation to keep going despite being interested in the topic.

3. HuggingFace NLP Course: To complement my theoretical studies, I set out to complete the Hugging Face NLP course (https://huggingface.co/learn/nlp-course). This course showcases code examples (using the widely used HuggingFace library) alongside explanations. The course consists of nine detailed units, and my goal was to complete all nine to not only understand but also apply concepts such as tokenization and model training in real-world scenarios.

Weekly actions: Study daily (started with 30min/day, revised that to 15min/day) (yes, in addition to the reading…)
Results: Finished 6/9 units. Actually studied on only 33/84 days.
Mistakes I made: Made the same mistake as with daily reading. I could have massively reduced my stress by creating focus days with clearly defined hours for studying and achieved even more. The weekly actions were massively oversized in relation to the goal, leading to overwhelm.

Success: Despite talking so much about my mistakes, this was a massive success for me.

An overview of my goals in Health and Content Production

Here are my goals for “Health” and “Professional” (which in this case I interpreted as my side business – content creation). I’ll keep it short because the focus here is on the studying aspect above:

Health: Run 5k without walking breaks

My overall goal this year is to become the fittest version of myself with a mix of cardio and weight lifting.
Weekly actions: Go for a run 3x per week.
Results: Went for a run 0.83x per week. Managed to run 2.5km without walking.
Mistakes I made: I didn’t actually schedule the said exercise sessions.

Content Creation: Increase followers and watch hours

Last year, I barely posted content. This year, I want to give it a proper try and see what could happen. The goal, after some deliberation, was to reach 22k followers on Instagram (from 21k) and 3k watch hours on my YT channel (the observant among us will notice that these are two goals…)
Weekly actions: Post on YT once a week, Post on Instagram daily (so 7x per week)
Results: Posted 0 YT videos, did not increase my watch hours. Posted 2.75x per week on Instagram, did not significantly increase my follower numbers.
Mistakes I made: Again, I didn’t even schedule the work for YT. I never decided on video topics and never even started the work. Choosing follower numbers as a metric was also a bad idea. I went for the easy win by focusing on Instagram instead of YT, so I need to remove that option in the future.

Lessons Learned and “How I would do it better next time”

Adopt a “Sprint” Study Strategy – Maybe Agile Wasn’t the Wrong Idea After All

Setting unspecified daily reading times led to procrastination instead of forming a habit. For future quarters, I plan to specify exact study days and durations, potentially reducing study days to 3x/4x per week to allow for mental rest and variety. I noticed that I tend to work in sprints: some weeks I would study almost every day, then almost stop completely for 2 weeks. If I can incorporate this into my schedule, my output will probably increase.

The Importance of Revision When Studying

I underestimated how much content I would be reading and studying in 12 weeks (even without meeting my goal). I have some notes, but by unit 5, I already realized how much I had forgotten from earlier units. To avoid this lack of revision and reflection in my actions, I plan on integrating techniques like writing tutorials on newly learned topics to engage with the learned content.

Simplifying to Amplify: Yes, 6 Goals Were Too Many.

My experience taught me that overloading with sub-goals can indeed scatter my focus. For the next quarter, I will set fewer, clearer goals. Promise 😀 I need to focus on the most important goals, the ones that will make me truly sad if I haven’t achieved them after 12 weeks. Deciding what those are, however, is harder than it looks.

Tracking and Motivation: Notion and Ticking Things Off a List Is Great.

Using a table with all my weekly actions every week to track my progress was truly a game changer. Increasing the number of checkmarks within the week was all I cared about. This provided a visual and numerical representation of my commitment, which was highly motivating even when I fell short on some goals. The goal might have been already out of reach, but I could still make a difference in that week. I used Notion for this, but any table with colorful checkmarks works:

Conclusion: Sticking With It in Q2

The “12 Week Year” has proven to be a transformative framework for me, more so than any other productivity method I’ve tried. It offers a manageable way to implement significant changes and track progress effectively. As I refine my approach, I look forward to sharing more about my journey and the lessons I’ve learned along the way.

Stay connected for updates on my second-quarter goals and more insights by following me on Instagram. Bye 🙂

Leave a Reply

Consent Management Platform by Real Cookie Banner