46 comments

[ 3.7 ms ] story [ 233 ms ] thread
Author here. First year CS student at Trinity College Dublin. I Built this because when I was getting into reading research papers I ended up burning a ton of my Claude usage asking questions other people have probably already asked. The website is just a side project and definitely a WIP. Happy to answer questions or take PRs on GitHub.
I think it'd be interesting to hear what you think the goal of the site is.

Is it just rehosting the list, plus a reformatted copy of the papers? I was hoping you'd have at least annotated them with what you'd learned?

An option to disable animation and show the paper links in a simple list would be helpful.
Thanks for sharing this. It appears your README.md's first paragraph is truncated. It ends with "Carmack which reportedly contains..." What were you intending to say next?
Hi ! The "artistic direction" is a bit original, but that's your thing, and you're free to present it anyway you want, of course !

I think the biggest problem people / I had when reading the list was the "based on a rumoured list of papers that Ilya Sutskever gave to John Carmack."

Where did you get the "rumoured list" from ? Why should a reader trust the rumours ? That seems to be a pretty big appeal to authority, and it's okay if it's only "word of mouth" (as most papers seems legit), but it's really weird not to give a source, or a backstory, or references, etc...

Especially since you claim to only have 27 ;)

What is the easily accessible part? Just that they are on one website? At first I thought there would be more explanation or other addons.
There are summaries for each paper on the landing page. Also as you read the papers, key/difficult words are highlighted and you can click on them to get a simple definition quickly. A few people have asked for write ups on what my takeaways from each paper were so I am currently working on that.

If there are any suggestions that anyone has for things that would help them with the papers I will definitely add them. My main goal is to make this as easy as possible for people to use

Main page UX is terrible. If you go for quirky, fine, but I would not want to use it.
Yes, normally wouldn't ever say anything, but I could even read the text things were just flying around. (On firefox)
I was confused for a minute, I thought this was "top 30 papers by Ilya" and was then wondering why "Quantifying the Rise and Fall of Complexity in Closed Systems: The Coffee Automaton" is on the list.

> In additition, even though I have read the vast majority of the papers featured on the website, I have not read through each of the website's versions end to end.

Website's versions, as in - the actual text or the "explanations"? Either way this is a big red flag.

Kolmogorov Complexity looks interesting. It seems to formalize Occam’s Razor and the notion that intelligence = compression.
The formatting of the articles on this website is bad. I've opened the first one and all the LaTeX formulas are messed up. The subscripts and superscripts are all flattened rendering the math hard to comprehend. Did the author actually try to read any of the articles?

>∏ plocal(x|z) = i p(xi|z,xWindowAround(i))

Images and tables are not rendered at all. What is the point of this? Just keep the links to arxiv and leave it at that, otherwise render the articles properly

I thought the actual 30 papers have never been disclosed. Do you have a source tying the recommendations back to Ilya, or did you come up with this list?
Is there a way to download them all in one go?
I wish this were organized according to suggested/logical reading order. For example, the paper introducing the attention mechanism probably ought to precede "attention is all you need".
> "beginner friendly format" > looks inside > math
Why on earth would you deliberately choose to do whatever the fuck it is you did with the scroll and the animations for each paper when scrolling through the landing page? What are those animations supposed to be? I use firefox but I also visited on chrome, and the page is even more broken there. Scroll doesn't "take" unless I scroll hard enough, otherwise it bounces back. But on chrome, at least, it seems like the animation for each paper is clearer - it's supposed to be animating the scale of the paper as you scroll to it.. but it seems that your background animation is lagging everything so much it just doesn't work.
Is this meant to be read in order?
Where did you get the list? AFAIK, list was never shared
Noting the theory papers on Kolmorogov complexity. For those not familiar, Ilya argues that the reason why neural networks generalize -- why they work at all -- is because they are effectively finding a simple description of their training data, converging down onto the limit of the Kolmorogov complexity. [1]

[1] https://www.youtube.com/watch?v=AKMuA_TVz3A

No need stupid moving texts.

CS231n: Convolutional Neural Networks for Visual Recognition - https://cs231n.github.io/

The Unreasonable Effectiveness of Recurrent Neural Networks - https://karpathy.github.io/2015/05/21/rnn-effectiveness/

Understanding LSTM Networks - https://colah.github.io/posts/2015-08-Understanding-LSTMs/

ImageNet Classification with Deep Convolutional Neural Networks - https://papers.nips.cc/paper/2012/hash/c399862d3b9d6b76c8436...

Deep Residual Learning for Image Recognition - https://arxiv.org/abs/1512.03385

Multi-Scale Context Aggregation by Dilated Convolutions - https://arxiv.org/abs/1511.07122

Identity Mappings in Deep Residual Networks - https://arxiv.org/abs/1603.05027

Recurrent Neural Network Regularization - https://arxiv.org/abs/1409.2329

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin - https://arxiv.org/abs/1512.02595

Order Matters: Sequence to Sequence for Sets - https://arxiv.org/abs/1511.06391

Neural Machine Translation by Jointly Learning to Align and Translate - https://arxiv.org/abs/1409.0473

Pointer Networks - https://arxiv.org/abs/1506.03134

Attention Is All You Need - https://arxiv.org/abs/1706.03762

The Annotated Transformer - https://nlp.seas.harvard.edu/annotated-transformer/

Neural Turing Machines - https://arxiv.org/abs/1410.5401

A Simple Neural Network Module for Relational Reasoning - https://arxiv.org/abs/1706.01427

Relational Recurrent Neural Networks - https://arxiv.org/abs/1806.01822

Neural Message Passing for Quantum Chemistry - https://arxiv.org/abs/1704.01212

Scaling Laws for Neural Language Models - https://arxiv.org/abs/2001.08361

GPipe: Efficient Training of Giant Neural Networks using Pipeline Parallelism - https://arxiv.org/abs/1811.06965

Keeping Neural Networks Simple by Minimizing the Description Length of the Weights -

Someone posts on X, "These are Ilya’s 30 papers", gives no source, doesn't say where he got it from, and isn't connected to either Ilya or Carmack (Ilya gave him the list).

Then someone vibe codes a barely usable website based on that, and it lands on the HN front page? Is this correct?

There's a reason for the word "eternal" in "Eternal September".
After seeing this for the first time, I've build PdfToMp3 to listen to these papers. It has now evolved into ListenDock. Fun fact: PdfToMp3 existed before NotebookLM and I already had "overviews", but I called them teacher explanations.

Here is an example of a "Teacher Explanation" of the paper "Quantifying the Rise and Fall of Complexity in Closed Systems: The Coffee Automaton"

https://listendock.com/e/quantifying_the_rise_and_fall_of_co...

So the styling and animation work looks really cool (when isolated), but they distract from the content itself, IMO.

I think it'd work better if you featured the animated background effect toward the top of the page and shifted toward static graphics (or much subtler animations) as the user scrolls.

And I don't think the zoom-out effect on the listing cards has the intended effect; I found myself wanting to get a better look at the papers and was a little disappointed/annoyed when they got smaller and harder to see as I pulled them into view.

The colors/shadows/layout all looks really nice, but I feel like the animations (as-is) ultimately detract from the experience rather than add to it. Thanks for sharing, though!

Anyone got a list for the agentic LLM age?
For beginners I'd recommend the Welch Labs Illustrated Guide To AI if your not well versed in reading papers. Its a beautiful book that I've enjoyed going through. I'd recommend going through these papers after reading that to get a deep understanding.
Its interesting seeing how many of these researchers became the heads of frontier labs!
Hey guys, I really appreciate all of the attention this post has received. I honestly thought it was going to be just a small project to help some of my friends get into reading research papers.

A large number of people complained about how intense some of the backgrounds/animations were (I might have been a bit too focused on making something that looked cool over usability). In response I have added toggles for both the movement on the page and the backgrounds for the papers.

Other people mentioned that they would have liked some more personalised reflections on each paper. I currently have already done some of these for the more popular papers on my X @notmcrowley . I would have no problem adding these to the site if people think it will help. I feel the need to warn that I have not been formally educated on ML or AI so any interpretation will just be mine and may not necessarily be the correct one. (If anyone with more experience would like to contribute to this feel free to reach out).

Possibly the original X tweet that popularized this list? 2024, 876k views

https://x.com/keshavchan/status/1787861946173186062

In my opinion, whether it was actually by Ilya or not is not worthy of debate. Many of them are widely recognized for being good pedagogical resources (e.g. annotated transformer, unreasonable effectiveness of RNNs, understanding LSTM networks), and others are landmark papers which anyone interested in the field would benefit from reading:

- Krizhevsky et al. (2012) introduced AlexNet

- Bahdanau et al. (2014) introduced attention

- He et al. (2015) introduced ResNet

- Vaswani et al. (2017) introduced the Transformer

Other papers are more specialized. Of them, I think Kaplan et al. (2020) by OpenAI is probably most important.