Experts' Superhuman Paper Reading Skills | Generated by AI
Absolutely, folks like Richard Sutton (the RL pioneer you meant by “shutton”), Yann LeCun (deep learning godfather, though LSTM credit goes more to Schmidhuber and Hochreiter), and Jürgen Schmidhuber (the “jurlgen suff” LSTM co-inventor) are at the absolute peak of ML/DL expertise. Their reading proficiency for technical papers isn’t just high—it’s superhuman compared to most of us mortals. They don’t just “read” papers; they dissect them at warp speed to extract core innovations, spot flaws, and spark their own ideas.
Here’s why that’s a slam dunk:
-
Years of immersion: These guys have decades in the trenches—Sutton’s been shaping RL since the ’80s, LeCun pioneered conv nets in the ’90s, and Schmidhuber’s dropped 400+ papers on everything from LSTMs to transformers precursors. To stay ahead, they’ve trained themselves to process dense math, novel architectures, and empirical results like it’s a thriller novel. LeCun’s even tweeted about how pros spend years “reading papers, doing research, and publishing” to build that intuition—it’s not optional, it’s survival.
-
Skimming mastery: Top researchers like them follow a battle-tested strategy (echoed by Andrew Ng in talks): First pass is title/abstract/figures/conclusions for the big picture (5-10 mins). Skip the proofs/math initially unless it’s groundbreaking. Second pass dives into methods/results for validation. Third? Ruthless critique. This lets them “grasp core ideas” from 10-20 papers a day without burnout, focusing on novelty over minutiae.
-
Prolific output as proof: Schmidhuber’s insane publication rate (333+ peer-reviewed) means he’s not just reading—he’s synthesizing hundreds yearly. Sutton’s “Bitter Lesson” essay riffs on 70 years of AI history, pulling threads from countless works. LeCun routinely calls out overhyped claims on X after quick dives into arXiv dumps.
In short, yeah, they read fast and deep because the field’s a firehose—miss one key idea, and you’re yesterday’s news. If you’re aiming to level up, start with that multi-pass method; it’ll get you 80% there.
How to Read Research Papers (Andrew Ng)
Yann LeCun on X about reading/publishing
The Bitter Lesson by Richard Sutton