Resources that helped

I have been studying ML and particularly LLMs for more than half a year already. This is a reflection of what I found helpful. Sometimes doing something I asked myself, ‘Is it indeed that thing that will give me a desired boost?" Below are the resources that gave. This list might be changing once in a while, but everything is changing. I also do not want to provide huge lists of sources, books, and lectures; it will be concise, but dense.

List

Block 1. Math and context

1. 3B1B. Classic, but it's really the best
   • Essence of linear algebra
   • Essence of calculus
   • Neural Networks
2. Visualization of NanoGPT with description
3. State of Al report and tech tendencies. Super important for context I was just sitting and searching for every technique and company mentioned there.

Block 2. People and context again

4. Cohere Labs Discord. Community of researchers that conducts a lot of events explaining the latest research.
5. X. Full of information, companies, and people from field there.

Block 3. Fundamentals

6. Andrery Karpathy videos. You can start implementing your transformer and BPE tokenizer.
7. Reading fundamental papers.
8. Hardware. Partucularly CPU and GPU, and others are just a very good videos as well.

Block 4. Practice

9. CS336. If you will be able to complete all the assignments, it will be a very strong start. Lectures are full of non-trivial and up-to-date details.
10. Explore open source repositories (whatever you find interesting: vllm, Pytorch, Transformers, etc…)

---

To be honest, I was surprised how wide the field is. I would not skip any of this steps, only extend the list. At this point I strongly believe there are no shortcuts here and you need to learn the details.

After web development, ML is a big shift. I found it helpful to have an ML pipeline visualization in my head for seeing the whole picture.

Forming the right mindset is another thing: particularly always have efficiency, compute, time, and memory in mind. Also, the model should be accurate.

The field is moving forward very fast, so checking the latest updates is an important thing, preferably every day, just to keep track. Any source is good: podcasts, X, blogs, and some new papers. It's definitely impossible to read everything, but it allows you to have a good understanding of what is going on.

Choosing the direction. Without it, it's a big chance to be lost in algorithms, information, papers, and optimizations.

While this collection of resources helped shape my understanding of ML, your optimal learning path will likely look different. Use this as a starting point to explore what resonates with you, adjust based on your background. I've experimented with various learning techniques and productivity tools, but ultimately discovered that quality resources + finding my own rhythm matters most. That's why I'm sharing these specific materials. Your path through them will be uniquely yours.