How I Taught Myself to Critique AI Research

How I refined my articles without having anyone to critique them.

8 min readMar 10, 2024

Before we begin, RIP to Akira Toriyama. Great art moves us spiritually, and DBZ is no exception. I hope a small part of his legacy is honored through the work I put into creating beautiful articles.

Writing has been the best career decision I’ve made in my life. It’s given me exposure to new potential clients, acts as an additional source of revenue, and allows me to stay updated on the most important developments in the industry. Even if you don’t explicitly want to be a writer/publish for the public, writing for yourself can help with a few benefits-

It acts as a store of knowledge so that you can refer to things later.
Writing things down allows you to spot your own cognitive blindspots and understand where you might be diverging from the answer. This is critical in getting to a high level relatively quickly.
Writing about a source/topic forces you to engage with it more deeply, helping you understand it a level that’s more profound than just reading about it. This is a variant of the Feynmann Principle- you understand something better by teaching it.

These benefits are one of the reasons why I ask the clients whom I work with for interview preparation to take diligent notes as they solve problems. It allows us to spot their weaknesses and game plan for their individual tendencies and weaknesses.

When I started writing, I had to overcome a few challenges:

Being self-taught I didn’t have the same context as an ‘educated’ person who would have a clearer understanding of what was important to academics (the people I wanted to write to impress). This becomes doubly clear with AI Research papers, where there are some great resources for intro-level information (“what is a neural network etc”) but not so much for the cutting-edge stuff.
I also didn’t have a rigorous or well-defined learning path- which made it hard for me to understand the gaps in knowledge and to work on what I didn’t know (especially with my unknown unknowns).
I didn’t have a peer group/network that could critically evaluate my work and give me feedback. No one in my circles interacted with Machine Learning Research, so I had no one to tell me if I was on the right track or what mistakes my analysis contained.

In this article, we will be looking at how I addressed those challenges as I was getting started. I hope the way I did it can help all of you in your journeys. I applied similar game plans (with some tweaks) to developing my Leetcode Solutions, Understanding System Design, and other important concepts, so I’m sure this game-plan will help you a ton.

Join 150K+ tech leaders and get insights on the most important ideas in AI straight to your inbox through my free newsletter- AI Made Simple

Steps that I took to Improve my AI Analysis

Find an Ideal: As they say, “Imitation is the sincerest form of flattery that mediocrity can pay to greatness.” When starting, you will not be good at writing. It’s important to have a high-quality reference that you can hold your work to. When I started, that was Yannic Kilcher, one of AI most high-profile researchers and content creators. I would pick research papers that he had already broken down, and write my analysis based on the same paper. My process was simple:

Pick a Yannic Video that looked interesting (don’t watch it yet) and didn’t have too much controversy (people weren’t disputing the results or throwing around too many counter-arguments). This was very important b/c I didn’t have the confidence in my skills and my knowledge still had a LOT of flaws. I could not jump into the more complex stuff and I didn’t want to write about something I didn’t understand.
Read through the paper to see how much I could understand
If I could understand the details, then I would try to break it down.
After breaking it down, I would watch Yannic’s breakdown to see how much I understood. I would take notes of all the points I missed.

The comparison had two major benefits- it allowed me to spot both flaws in my knowledge AND improved the quality of my explanations. If Yannic said something that wasn’t immediately obvious to me, I knew what I had to study next. Most importantly, if my analysis overlapped with Yannic’s, I knew that I had the right thing.

The important thing here isn’t to plagiarize your reference but to use that to support yourself. Compare Yannic’s video to my first-ever AI Research article here. Even though it’s on the same topic, the result ends up being very different. That’s because I chose to omit certain things that he talked about and add a few of my own. This is key because trying to copy a reference too faithfully will only lead to your work being a cheap imitation, and will not provide any of the personalization benefits. Ultimately your writing should reflect your judgement and your knowledge, no matter how incomplete or flawed. This is what eventually allowed me to take a path and develop a style that was very different to other AI Writers that touch similar themes/topics.

This first step will help you a lot with the knowledge part of writing (providing you with some guidance on what you should study). Comparing divergence with your source can also be a good substitute for peer feedback (how much of this video did my analysis miss?). The other principles we discuss will focus on how you can actually get your writing finished. When you’re a begineer at writing, it can be difficult to know when you should stop a piece, which is why you should narrowly define the scope of your work before/as you create it.

Pick a small subset:

One of the biggest problems with writing is that it is extremely difficult. It’s much easier to talk at length about a certain idea than it is to produce a deep-dive writeup on that topic (this was my motivation for the YouTube channel). Therefore, don’t write for the sake of writing, and don’t try to write down everything. Define the scope of your writeup and stick to it mercilessly. Trying to cover too much would be too difficult, especially when it comes to threading narratives b/w them. This is true for me even today. Every article you read is completed on the back of 10 related articles that were discarded.

For my articles, I would define exactly what the article would teach readers before writing, so I could focus on that. For the interview prep squad, our focus was always on where their approach went wrong and they started missing the solution (whether it was a lack of knowledge, identifying the correct approach, execution, or communication). The following technique is very helpful to limit scope.

Blackbox a lot

To cover the benefits of blackboxing for writing, it’s good to first understand why it’s such a powerful idea for learning difficult things. Learning things from scratch is a waste of time. Especially in an always-changing field like Tech. Instead, when interfacing with complex topics, it’s best to follow a black-box approach. We covered black-boxing here, in the section on how to study Data Structures and Algorithms for Leetcode

The first step is to black box the DS/A. Simply put understand two things- the inputs it takes and the outputs it generates. Understand this stage intimately- if someone asks you to find a value in a sorted list you should scream binary search, even if you can’t implement the binary search yet.

If you can’t understand something, skip it for now. Come back to it later. Build your familiarity with the idea slowly. If your soul completely rejects understanding something, that’s a good time to use a tutor to get deeper insight into it. The mistake that a lot of beginners make is that they try to understand the complete idea at the very beginning. This will overload your brain and leave you with more questions than answers. It’s better to mark something as a black box and return to it incrementally.

When it comes to writing, it is very easy to get distracted by the details and to dive into tangents. Black-boxing is great for helping you limit the scope of your writeup, which will stop you from wasting time and let you complete your work.

By blackboxing you make implicit assumptions about your audience and their knowledge and their interests. This is where the last advice can be very helpful.

Write for yourself

The best standard for quality is to always think of the following: “Is my writeup useful to me, as I am now“. This is obviously good if you’re writing just for yourself, but will also be very useful to you if you want to write for other people. In my mind, if my article would have helped me understand a research paper better, it would probably help others as well. This can be a fuzzy benchmark for ‘informativeness’ b/c cognitive blindspots can skew your assessment, but it’s still worth looking at all your work through that lens. It can be really good for evaluating how readable your writing is.

These were the guiding principles I used to write my first few articles. As I published more, I started getting some attention online, which allowed me to refine my work with the feedback of other people.

If you liked this article and wish to share it, please refer to the following guidelines.

That is it for this piece. I appreciate your time. As always, if you’re interested in working with me or checking out my other work, my links will be at the end of this email/post. And if you found value in this write-up, I would appreciate you sharing it with more people. It is word-of-mouth referrals like yours that help me grow.

If you find AI Made Simple useful and would like to support my writing- please consider becoming a premium member of my cult by subscribing below. Subscribing gives you access to a lot more content and enables me to continue writing. This will cost you 400 INR (5 USD) monthly or 4000 INR (50 USD) per year and comes with a 60-day, complete refund policy. Understand the newest developments and develop your understanding of the most important ideas, all for the price of a cup of coffee.

Become a premium member

I regularly share mini-updates on what I read on the Microblogging sites X(https://twitter.com/Machine01776819), Threads(https://www.threads.net/@iseethings404), and TikTok(https://www.tiktok.com/@devansh_ai_made_simple)- so follow me there if you’re interested in keeping up with my learnings.