Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
30papers.com – Ilya's 30 essential ML papers, in a beginner friendly format (30papers.com)
551 points by notmcrowley 20 hours ago | hide | past | favorite | 85 comments
 help



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?


First year CS student excited to learn about a thing puts together a small website of academic papers, posts it to HN to share with others.

Then someone makes a shitty comment. Is that correct?


It's always fun to remember how rude people were about Dropbox on here.

More seriously though, in many ways HN is a pretty broad church. You're always going to get a spectrum of opinions, and some portion of people are always going to be particularly forceful about putting forth their opinion. Maybe it's a bad day, maybe it's habit, varies from person to person and day to day, etc.

But I think, if you're posting something to HN (or, really, any large internet forum), negative feedback - and dismissive feedback - is something you need to be prepared for as part and parcel of the experience because it often is going to happen.

Not that I agree with the person you're responding to - their remark struck me as quite a mean-spirited and unnecessary comment, and I very much prefer your perspective.

Anyway, I've bookmarked the site so make of that what you will.


It's hard sometimes because say it's something that I wrote but someone else posted to HN, I've just had a lot of people's opinions foisted on me.

I'm relatively immune to a lot of things, but we're also entering a world where a lot of people can build and might not expect to have potentially millions of people critiquing their work to the level they do.


We are way past that point now. It's a common tendency around here, especially taking into account how often those type of posts are actually the ones on the top. Gladly we still have people like your parent to call them out immediately.

That's just hn for you. Not that it's a good thing (as per me at least), but that's what hn is, no matter how much it (or few from here) tries to think/pretend otherwise.

Shitty comment?

From my point of view the Jedi are Evil


First year CS student is already vibe coding something which lacks basic presentation skills.

If he was really excited he would have put an effort and he wouldn't clickbate submission title.

But alas here we are, perhaps we should give him particpation star for effort.


First year CD student excited to learn puts together a website, and more experienced guys makes a shitty comment that puts things into context. Then someone makes a funny comment mimicking its structure. Is that correct?

No. It's quality control. This is a classic clickbait formula headline with 0 backing. I just typed "list 30 papers to get me started in machine learning" into an llm and got 27/30 of these...

This page is an LLM prompt response as a list of jpegs with a fake title. You can probably just add "and prepare it as a webpage with image previews for each" ...

I think we can do better than someone shitposting a sentence into openclaw and getting it to the frontpage.

People on here are actually building 100 billion dollar companies, publishing in prestigious journals, maintaining transcontinental infrastructures with global reach... let's hear from them instead of a mac mini going burr for 15 seconds.


It actually got to the front page though? I think what's more ironic is that the top comment thread on this is devolving into a discussion about hacker news meta and nobody in the comments section has written even 1 line about what these papers are or the subject or how useful they/the website is. Yes that is correct.

Many here mistake passive aggressiveness for civility. So does the rule enforcement.

100% correct (anyone on hn for longer than 3 months will recognize that this is exactly the culture here).

not shitty enough

Compiled resources for nerds are catnip. Hit that bookmark/upvote button to never get to it :)

I feel seen. Straight to the "stuff to read later" bookmark pile/mausoleum :)

One time my brother upgraded his RAM so he could keep more tabs open. I think he was up to a couple hundred.

Rookie numbers! You gotta pump those numbers up!

I wish this was more wrong.

attention is all you need

Useful list, but just needs an awesome list with links: https://github.com/ianchanning/ilya-sutskever-recommended-re...

> anything that gratifies one's intellectual curiosity

close enough imo

any apparently 400 voters too.


they kind of mention the source on their website though

" rumoured list of papers that Ilya Sutskever gave to John Carmack. "

there is aslo manning book called illya list

https://www.manning.com/books/sutskevers-list


Ah, another naysayer as if there is a scarcity of them

Delete the two options in the upper right hand corner of the homepage and Bob's your uncle.

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).


Please add them on the site for those of us who have never had Twitter and don’t plan to open one ever. Thanks for this compilation, I am — like your friends — trying to get into reading research papers and this is right up my alley right now.

Even with the motion/background button toggles you are still left with tiny fonts that make it hard to read.

It actually made me check my browser wasn't set to zoom out. But then using zoom changes nothing, which breaks accessibility.

Also why does the header need to take up 3/4 of the screen?

[edit] Clicking on a paper doesn't even bring the paper up. I have to hit another click to get to it.


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.

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?

An option to disable animation and show the paper links in a simple list would be helpful.

The problem is the background is often times doing a wave motion across the screen.

Then the foreground content is doing an in/out undulation on top. So you’re seeing an undulating in/out in every possible direction + the background. And the foreground animations are all at the same time. So it’s not that we’re emphasizing any one thing. We’re emphasizing all of it.

The key with animations is in what they’re trying to draw attention to, the character of the movement, and the timing of it. You usually don’t want everything to equally animate at once.

I would: • Use background movement that also isn’t a “wave” • Stagger the timing of foreground animations so the main content is emphasized, followed by a pause, followed by the sidebars • Change the nature of the animations so they’re not doing the same essentially thing “zoom and pan” - so have the center zoom and pan, but do something different for the sides.


Agree on the animation.

As an aside, I've seen folks mention respecting reduced animation hints and such in the past and was always curious about this because I've never had any negative experiences with animations... until now!

Something about the animations on this site did my brain in while scrolling through the papers, and now I "get it."


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?


Hey thanks for checking out the website. I did not expect this to get as much attention as it did. I was honestly just planning on having it as a small side project for my friends and some others who would like to get into this kind of stuff. I will definitely annotate them in the future if that is something that people would appreciate. I currently have something like this done for a few papers on my X account.

I mean, you don't have to annotate them :) Would I love to read annotations? Sure - but it's a boatload of work for you, and if it's just meant as a repo for you and your friends, no need to do a ton of work just because an Internet rando asked.

But given the sudden wide audience, a quick "here's what this is for" at the top might be helpful.


> I think it'd be interesting to hear what you think the goal of the site is.

why do you care? this is a disingenuous question.


Has anyone tried just asking John Carmack...?

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".

Second this! And if the papers are in "logical reading order", it would be very useful if this is stated on top!

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...


Why do I get downvoted whenever I post something here? Do you think its too spammy? Because its AI? Do I have a downvote bot following me?

I haven’t looked at your other comments but the answer is your comment isn’t valuable.

Text to speech summarizing is a dime a dozen. Your audience here prefers reading a blog and is already annoyed by ai vs written by a human content so what you are offering is the opposite of what they want.


Ok, thanks for the viewpoint, that makes sense. I use AI summaries every day and find it very valuable. But I also see the trend e.g. on Reddit, that people are very dismissive of ai content.

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.

Bought because of this comment. Thanks!

always a fun day when someone rediscovers these. Lucky 10000

in case folks are interested, i wrote up a ~layman's review of each paper over the course of two years a while back. Several of those reviews ended up doing reasonably well on hn. Full analysis of the ~23 docs that were papers and not massive books

https://12gramsofcarbon.com/p/ilyas-30-papers-to-carmack-tab...


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


That's true of all statistical models, it's not some magic property of neural networks.

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.


Even if Ilya didn't really create this list I have a very good opinion about every paper on this page that I've read (most of them) so I think it's a great resource. Lately during my off time I want to do something related to AI research (which I am already doing full time atm so I need something light) and I am for sure going to read through this.

Nice presentation of the list!

I'd recommend watching a few of his talks/podcasts before during reading these to get the overview and how all the bits in these works tie together.

https://www.dwarkesh.com/p/ilya-sutskever

https://simons.berkeley.edu/talks/ilya-sutskever-openai-2023...

https://www.dwarkesh.com/p/ilya-sutskever-2


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.


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 - https://www.cs.toronto.edu/~hinton/absps/colt93.pdf

A Tutorial Introduction to the Minimum Description Length Principle - https://arxiv.org/abs/math/0406077

The First Law of Complexodynamics - https://scottaaronson.blog/?p=762

Quantifying the Rise and Fall of Complexity in Closed Systems: The Coffee Automaton - https://arxiv.org/abs/1405.6903

Kolmogorov Complexity - https://onlinelibrary.wiley.com/doi/book/10.1002/047174882X

Variational Lossy Autoencoder - https://arxiv.org/abs/1611.02731

Machine Super Intelligence - https://www.vetta.org/documents/Machine_Super_Intelligence.p...


Upvoted. Did you compile that list just now, pulled it from bookmarks, or other source?

this is a goldmine, worth bookmarking.

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?

I think someone on Twitter made it up. It was also 40 papers, not 30. https://dallasinnovates.com/exclusive-qa-john-carmacks-diffe...

The list I got was from ex-OpenAI employee Andrew Carr on X. I believe he said in his post however that the list he uploaded is not the full list they were provided at OpenAI however.

This list was made by some guy on twitter. https://x.com/keshavchan/status/1787861946173186062

It's unknown whether it has anything to do with Ilya Sutskever.


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!



Is there a way to download them all in one go?


  for x in 1611.02731 1511.06391 1811.06965 1512.03385 1511.07122 1704.01212 1409.2329 1512.02595 1706.01427 1410.5401 1806.01822 1706.03762 1409.0473 1506.03134 2001.08361 1405.6903 1603.05027 math/0406077; do curl -fL https://arxiv.org/pdf/$x -o ${x##*/}.pdf; done
  for u in https://www.cs.toronto.edu/~hinton/absps/colt93.pdf https://proceedings.neurips.cc/paper_files/paper/2012/file/c399862d3b9d6b76c8436e924a68c45b-Paper.pdf https://www.vetta.org/documents/Machine_Super_Intelligence.pdf https://www.lirmm.fr/~ashen/kolmbook-eng-scan.pdf https://scottaaronson.blog/?p=762 https://karpathy.github.io/2015/05/21/rnn-effectiveness/ https://colah.github.io/posts/2015-08-Understanding-LSTMs/ https://nlp.seas.harvard.edu/annotated-transformer/ https://cs231n.github.io/; do curl -fLO "$u"; done


Kolmogorov Complexity looks interesting. It seems to formalize Occam’s Razor and the notion that intelligence = compression.

If you find this interesting, you should look into Solomonoff induction. It combines Kolmogorov complexity with Bayes rule to provide a general framework for inductive inference, and naturally formalizes Occam's razor.

I wouldn't say so about Occam's Razor which is a heuristic.

The relationship between compression and intelligence, while not equal is definitely there. It looks like 3Blue1Brown is going to be doing some videos on this aspect.


there's a way to connect kolmogorov complexity and occam's razor, which is solomonoff induction

Is this meant to be read in order?

This is a beautiful way to present extremely high quality information. I sometimes lament the unpleasant friction involved in finding and reading academic papers (the overly formal style is a necessary evil, but the irritating paywalls, followed by inevitable searches for '%{title} filetype:pdf' feel like unnecessary ones).

How is this a beginner friendly format?

Its interesting seeing how many of these researchers became the heads of frontier labs!

the format is not friendly at all...

Where did you get the list? AFAIK, list was never shared

Anyone got a list for the agentic LLM age?

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


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.

Myspace and 5th grader PowerPoint presentations had a vibe coded child.

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)

Indeed. I scoffed at your comment and went to the website. After scrolling a bit, I find myself having a mild headache and slight dizziness.

I would request the author to consider something that does not distract us from this educational and informative website ( I have bookmarked it ).


Indeed. It's very bad.

> "beginner friendly format" > looks inside > math



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: