The Story of a Lazy College Student with a 2.4 GPA Who Got Hooked on AI
Hello, I usually go by the name Lee-Pippi and am active on Twitter.
By the way, I recently changed my icon. My girlfriend drew it for me; the only things that look like me are the hairstyle and the fact that one eye is double-lidded...
I've been wanting to try writing a tech blog, so as practice, I've decided to post a note looking back on my year.
As a result of a 2.4 GPA student like me getting hooked on AI for a year, I am now participating in long-term AI-related internships at three different companies.
Myself before getting hooked on AI (2021-2023)
A serious club member (I was in a sports club and would skip classes to practice all the time).
I worked as a private tutor for eight students, and I would spend all the money I saved on traveling during spring and summer breaks.
By my fourth year, it's normal for others to have finished their credits, but I was about 14 credits short and graduation was looking a bit precarious. (I had failed required courses in my second and third years, so if I failed them again, I would be held back immediately; graduation isn't guaranteed yet...)
In other words, I was a typical lazy college student.
How I got hooked on AI (Early January 2024)
Since last winter, I had mysteriously become obsessed with reading books in fields I knew absolutely nothing about, buying and devouring books on stocks, mahjong, and shogi on Mercari.
Following that trend, I bought and read a book called 'First Deep Learning' out of curiosity. At the time, I was just coasting through my university classes, so I had zero knowledge of linear algebra, calculus, or statistics, and I could only understand about 20% of it...
At that time, I simply liked the cool-sounding term 'Deep Learning,' and that started my days of devouring AI-related books.
Also, around this time, I broke my leg during club practice, and it was confirmed that I would be out for three months and couldn't participate in any tournaments, so I was looking for something to get hooked on in my despair.
I don't think I would have gotten this into AI if it weren't for that injury.
The period of devouring AI books (January-March 2024)
My club schedule became completely free, and since I had a broken leg, I couldn't go outside and was completely bored at home, so I started a period of randomly devouring AI-related books while watching anime and movies.
Among the many things I read, I'll list the book I recommend the most. I bought O'Reilly books, but during this period, I couldn't understand the matrices in linear algebra and gave up immediately.
'Clear Understanding: Machine Learning with Python' is seriously highly recommended for people who have just started studying AI!
Reading this, I was able to understand the meanings of words I had vaguely thought were cool, and it was incredibly fun.
At this time, I didn't even know the concept of a Python library, so I thought numpy and pandas were the names of statistics textbooks, haha.
Words I thought were cool at the time:
numpy, pandas, matplotlib, Support Vector Machine, Random Forest
The GCI obsession period (April-July 2024)
Kaggle setback
I think it's a common experience for AI beginners to start trying their hand at Kaggle once they get a little used to machine learning and pandas.
That was exactly me; I was enthusiastic about joining a Kaggle competition and winning a medal, but I was soundly defeated. Even looking at the notebooks, I had no idea what was written. (Thinking back now, most of the notebooks at that time were written in polars instead of pandas, which was a high hurdle for me back then...)
At this time, I learned that the Titanic competition was a gateway for beginners, and I was happy to be able to post my first score on Kaggle by copying the code from a blog.
Taking GCI
I learned about GCI from a blog I was reading at the time and decided to take it. At that time, I was full of motivation and was fully intending to become one of the top performers in GCI.
I had researched GCI quite a bit, and before taking it, I understood that there were two competitions and what seemed to be an incredibly heavy final project.
What I did during the GCI course
The GCI classes themselves were quite similar to 'Machine Learning with Python' which is easy to understand, so I mostly skimmed through the lectures. Instead, I poured a massive amount of time into the competitions. By reading blogs and other resources through the competitions at this time, I gained a tremendous amount of knowledge about machine learning. I also spent a significant amount of time on the final project, working on PowerPoint slides all day long in July.
Things I did outside of GCI
Once I started doing machine learning, I realized that statistics seemed necessary, so from mid-April, I started studying statistics with the goal of passing the Grade Pre-1 Statistics Certification. I read through 'Tokei Web' to mostly understand the material for the Grade 2 Statistics Certification, and for the Grade Pre-1, the workbook was too difficult, so I read several introductory books on multivariate analysis, Bayesian statistics, time series analysis, and probability distributions at the university library to get a general overview. However, once I knew the general overview, I suddenly lost the motivation to master it, and from there, I gave up on taking the Grade Pre-1 Statistics Certification exam. I hope to try again if I can find the time... I was doing this from mid-April to the end of May. I got too hooked on the competitions halfway through, which led to less study time, and I ended up quitting.
Besides this, I also dabbled in SQL, AWS, and linear algebra and calculus for my graduate school entrance exams at the end of August.
First Internship: Starting a Data Analysis Internship at a Key TV Station
There was a senior in my lab who was knowledgeable about machine learning, and he told me about a job opening, so I applied even though I didn't think I had a chance.
In the interview, I said something like, 'I worked really hard on the GCI competitions and achieved a decent ranking!' and I got accepted. (I think I was around 42nd in the first competition?)
In terms of work, it felt like something I could do as long as I could handle pandas in Python, so I think I was able to pass with that level of skill. If it had been a hardcore AI position, I don't think I would have been accepted 100%.
Graduate School Entrance Exam Period (July-August)
During this period, I was solely studying for the graduate school entrance exams. I had hardly learned anything properly at university, so I had to study everything from scratch.
I studied linear algebra, statistics, calculus, algorithms, and AI-related topics evenly. I think it was worth the effort because studying math properly at this time really paid off later on.
My most popular tweet ever; I was confused because my followers started increasing rapidly around this time.
Summer School in Parallel with Graduate School Entrance Exams
Since GCI was good, I impulsively started taking three summer school courses at the Matsuo Lab at the same time. At this point, thanks to my hard work on the competitions, I was quite confident in my knowledge of machine learning. However, I knew nothing about deep learning. The prerequisite for this summer school was having basic knowledge of deep learning, which I didn't meet at all, but I felt confident after GCI and applied anyway. As a result, I completed the generative AI and reinforcement learning courses and received certificates, but I quit the finance course halfway through.
I didn't understand anything even after listening to each lecture, so every time a lecture handout was released, I spent 5-6 hours asking ChatGPT questions and summarizing them on paper, repeating this process endlessly.
Looking back, I think this really deepened my understanding of deep learning, reinforcement learning, and generative models.
The Summer School Intensive Period and the Second Internship (September 2024)
After finishing my graduate school entrance exams, I dove straight into the summer school program.
I was so obsessed that I would come home at night after the post-exam celebration party and review the summer school materials. I couldn't have imagined myself doing this last year...
Around this time, the GCI results came back, and I was quite depressed that I didn't achieve excellence. Looking back, I don't think my final project was good enough to earn excellence, but at the time, I was confident.
It was during this time that I was invited to join EpicAI via Twitter. Since I had neither the skills nor the experience, I was truly happy and grateful for the invitation.
(Apparently, they invited me because my choice of books was good, haha.)
The Period of Getting Deeply Involved in EpicAI (October)
During the first month of my internship, I created three demos for the expo while receiving guidance from the person who invited me.
At first, I had no idea how to build them, but because they taught me so carefully, I was able to complete them.
Looking back, I think this experience was a huge step forward.
The Period of Focusing on Research and Club Activities (November)
Actually, I had resumed my club activities around April and had been doing them in parallel, but as retirement approached, I started practicing with more seriousness.
Also, in my research, I was able to achieve good results that seem likely to lead to a paper publication.
Furthermore, I was able to join a joint research internship at the Matsuo Lab using the things I created in October.
Looking Ahead
I got tired and sloppy toward the second half, but from now on, I want to give my all to the work right in front of me at my internship and absorb as much as I can.
My world has expanded and I have met many amazing people, so I want to grow to a level where I can stand shoulder to shoulder with them.