Comics that help non-experts vaguely understand research papers. Made by a non-expert.
Joined January 2025
- Tweets48
- Following190
- Followers38
- Likes45
Agreed.
🤖 Made with AI
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
images.nvidia.com/pdf/Open-W…
25
Inspired by @SebastienBubeck’s recent post, I made a comic featuring Sébastien and @sama explaining how an internal OpenAI model disproved a famous conjecture related to the Unit Distance Problem. 📊✨
Since it’s a comic, I added some creative exaggeration and simplified the math to make the story more fun and digestible.
Here’s a visual take on why this result feels like such an incredible milestone. Enjoy 👇
🤖 Made with AI
Huge congrats to @YichuanM on the Best Paper Award at #MLSys26! 🎉
To help everyone (even non-experts like me) easily understand this amazing paper, I put together an educational comic! 📚 (Just to be clear, I created/compiled it rather than drawing it by hand!)
Please note that there might be some simplifications or slight exaggerations to keep it fun and easy to read.
Paper: arxiv.org/abs/2506.08276
🤖 Made with AI
LEANN just won the Best Paper Award at #MLSys26 🥹
still processing this.
paper: arxiv.org/abs/2506.08276
repo: github.com/yichuan-w/LEANN
huge thanks to all the amazing collaborators, advisors, and open-source contributors who made this possible ❤️
1
1
188
I made a short webcomic about DSGym for everyday AI enthusiasts like me! 🎨
Huge congratulations to @FanNie1208 and the team on the ICML 2026 acceptance! 🎉
Hope this helps more people enjoy your awesome research!
(Note: Some details are simplified or exaggerated for the comic.
For exact and accurate info, please check out the full paper!👇)
📄 arxiv.org/abs/2601.16344
🤖 Made with AI
🚀 Excited to share that #DSGym has been accepted to ICML 2026!
DSGym is a holistic, unified framework for evaluating and training data science agents with standardized abstractions and a modular architecture for adding tasks, agent scaffolds, and tools.
In this work, we:
🔍 Show that existing data science benchmarks are vulnerable to shortcuts: agents can often solve tasks without using the actual data.
📊 Release DSGym-Tasks, a curated task ecosystem that standardizes and audits representative benchmarks, filters shortcut-solvable tasks, and expands coverage with new scientific tasks.
⚡ Use DSGym for execution-grounded trajectory synthesis: with only 2K samples, we train a 4B model that outperforms GPT-4o on standardized data analysis benchmarks.
📄 Paper: arxiv.org/abs/2601.16344
💻 Code: github.com/fannie1208/DSGym
🤗 Dataset: huggingface.co/DSGym
🧵👇
1
75
Just made a short comic based on this new AI paper from Google! 🎨✨
📝 Title: NEXUS: An Agentic Framework for Time Series Forecasting
🔗 Link: arxiv.org/abs/2605.14389
*Just a heads-up: I simplified and slightly exaggerated some details so that non-experts like me can easily grasp the core idea. Hope you enjoy getting the rough gist of it! 🍜💻
@GoogleAI @googlecloud @penn_state
🤖 Made with AI
1
50
Just read the incredible "Lighthouse Attention" paper by @NousResearch and decided to turn it into an easy-to-understand comic! 🎨✨
I made this for everyday AI enthusiasts like myself, so keep in mind there might be a bit of oversimplification or slight exaggeration to make the core concepts click! 😉
Huge shoutout to the amazing team behind this research: @bloc97_, @SubhoGhosh02, and @theemozilla! 🙌
Check out the original paper here: 🔗 arxiv.org/pdf/2605.06554
🤖 Made with AI
Today we release Lighthouse Attention, a selection-based hierarchical attention for long-context pre-training that delivers a 1.4-1.7× wall-clock speedup at 98K context.
It runs the same forward+backward pass ~17× faster than standard attention at 512K context on a single B200, without a custom sparse attention kernel, a straight-through estimator, or an auxiliary loss.
During training, queries, keys, and values are pooled symmetrically into a multi-resolution pyramid. We then score every pyramid heads, and a top-k cascade selects a small hierarchical dense sub-sequence, and after a sorting pass that enforces causality, we use standard attention for token mixing. A brief full attention resume at the end converts the checkpoint back into a competent dense-attention model.
Validated this using 530M parameter Llama-3 models across 50B tokens, with up to 1M-token benchmarks across 32 B200s under context parallelism.
The work on Lighthouse Attention was led by @bloc97_, @SubhoGhosh02, and @theemozilla.
2
5
791
Just made a short comic based on this AI paper! 🎨✨
📝 Title: SANA-WM: Efficient Minute-Scale World Modeling with Hybrid Linear Diffusion Transformer 🔗 Link: alphaxiv.org/abs/2605.15178
*Just a heads-up: I simplified and slightly exaggerated some details so that non-experts like me can easily grasp the core idea. Hope you enjoy getting the rough gist of it! 🍜💻
🤖 Made with AI
1
88
A short comic to explain the idea from the paper “IsoDDE : The Isomorphic Labs Drug Design Engine unlocks a new frontier beyond AlphaFold”.
⚠️ This comic simplifies and exaggerates some concepts for easier understanding and may contain inaccuracies.
For precise details, please refer to the original Blog
@IsomorphicLabs
Paper : isomorphiclabs.com/articles/…
1
1
232
