Interview, Fireside Chat
Luís and João Batalha: Fermat's Library and the Art of Studying Papers | Lex Fridman Podcast #209
- Platform Purpose & Mechanics: Fermat's Library is a platform designed to annotate academic papers, allowing users to add rich comments, explanations, and context directly to the margins of PDFs and LaTeX documents.
- Annotation Format: The interface supports both LaTeX and Markdown, enabling users to display complex equations and formatted text alongside paper content.
- Journal Club: A weekly feature where the founders and contributors release annotated papers across physics, computer science, and mathematics to make dense content more accessible.
- Margins: A free, open-source tool derived from the Journal Club software, allowing universities and individuals to host their own paper discussions and annotations.
- Librarian: A browser extension that overlays functionality onto the arXiv preprint repository, enabling commenting and easy reference extraction without leaving the arXiv page.
- Open Science Stance: The founders argue that scientific papers should be freely accessible to the public, criticizing the current model where journals charge paywalls for research funded by government or industry.
- Paywall Economics: It is noted that journals use unpaid labor from researchers for peer review while profiting from the sale of these papers, creating a system where universities pay to access work funded by their own taxpayers.
- Preprint Dominance: In fields like mathematics, physics, and computer science, preprints on arXiv are crucial because traditional peer review in journals can take up to three years to publish.
- BioRxiv History: A 1960s NIH experiment to share biology preprints was canceled due to pressure from the journal industry; currently, only 2% of biology papers appear as preprints compared to higher rates in math and CS.
- Annotation Example: Fermat's Last Theorem: The name "Fermat's Library" references Pierre de Fermat's famous margin note, symbolizing the potential energy trapped in the margins of scientific texts that annotations can release.
- Historical Context: The founders believe that understanding the "backstory" of scientific discoveries—such as the human struggles, random inspirations, and serendipitous moments—helps sediment knowledge and makes ideas more memorable.
- Feynman's QED Backstory: Richard Feynman's inspiration for quantum electrodynamics came from watching a waiter's plate wobble in a Cornell cafeteria, reigniting his interest in physics during a dark phase.
- Ian Goodfellow & GANs: Ian Goodfellow conceived the idea for Generative Adversarial Networks (GANs) in a bar, highlighting how some breakthroughs happen quickly in mature fields, though this is becoming rarer in machine learning due to the need for large-scale experiments.
- Dijkstra's Algorithm Origin: Edsger Dijkstra developed his shortest path algorithm in 20 minutes at an Amsterdam coffee shop without pen or paper, forcing him to rely on mental constraints that contributed to the algorithm's simplicity.
- Newton's Alchemy: Isaac Newton was deeply invested in alchemy and biblical prophecy, interests that are often omitted from popular science narratives but formed part of his broader intellectual framework.
- Freeman Dyson: Described as a model of humility, Dyson repeatedly expressed self-doubt regarding his qualifications to annotate papers despite his immense contributions to physics.
- Terence Tao's Blog: Mathematician Terence Tao is cited as an example of open science, sharing his research struggles and successes on his blog, which led to the solution of an Erdős problem via a reader's comment.
- Grigori Perelman: The mathematician who proved the Poincaré Conjecture published his proofs on arXiv as a protest against the traditional publishing system and later rejected both the Fields Medal and the $1 million Clay Prize due to disputes over credit and the nature of the scientific awards.
- Peer Review Critique: The current peer review system is described as broken because it relies on a small number of anonymous, unpaid reviewers, lacks incentives for rigor, and is skewed by the "tyranny of metrics" like impact factors.
- Impact Factor Distortion: Journals with high impact factors rely on "clickbait" titles and abstracts to sustain citation counts, turning scientific output into a "luxury good" economy and pressuring researchers to gamify their publication strategy.
- Crowdsourced Reviewing: The discussion proposes a shift toward crowdsourced peer review, where thousands of people can contribute small, specific insights, similar to Amazon reviews, to create more resilient and accurate assessments of scientific work.
- Startup Vision: Fermat's Library operates as a side project with a 20-year vision, prioritizing slow, organic growth over rapid scaling to avoid the pitfalls of the standard startup reward model.
- Wikipedia Analogy: The founders compare the potential for Wikipedia in science to its current existence on the internet, noting that a publicly edited, crowdsourced model can succeed where others predicted failure.
- Paper 2.0 Evolution: The speakers suggest the future of scientific communication may move beyond static PDFs to include multimedia elements like code, datasets, and video explanations to improve clarity.
- Reading Strategy: Feynman's reading hack of starting with the conclusion and attempting to reconstruct the logic in minutes is recommended as a way to bypass the density of traditional paper structures.
- Note-Taking Philosophy: The most effective notes are those written by one's "past self" for one's "future self," ideally capturing the specific struggle points and mental transitions made during learning.
- Video Integration: The founders are exploring audio and video formats for papers, citing channels like "Machine Learning Street Talk" and the "Code and Coffee" approach of George Hotz as examples of effective scientific explanation.
- Annotated Paper Example: The Samos Tunnel: A paper on the 6th-century BC tunnel on the island of Samos was annotated to correct a 2,000-year-old misconception about its construction, proposing that ancient engineers used a "rifle sight" method with sticks to align tunnels.
- Annotated Paper Example: Enrico Fermi's Trinity Test: Fermi's one-page, classified report calculating the energy of the first atomic bomb by dropping paper scraps was annotated to explain the "back-of-the-envelope" physics and applied to the 2020 Beirut explosion data.
- Annotated Paper Example: Bitcoin Whitepaper: The first paper on Fermat's Library was Satoshi Nakamoto's Bitcoin paper, annotated to explain foundational concepts like cryptography and Merkle trees for undergraduate-level readers.
- Annotated Paper Example: Dyson Sphere: Freeman Dyson's one-page paper proposing the concept of a star-harvesting sphere and using infrared signatures to detect extraterrestrial intelligence was featured to demonstrate the power of concise, visionary theoretical work.
- Game of Thrones Analysis: Luis Batala co-authored a paper using differential equations to determine the kill ratio required for the Army of the Living to defeat the Army of the Dead, illustrating the application of math to fictional scenarios.
- Soccer Analytics: The founders discuss the potential for computer vision to definitively compare Lionel Messi and Cristiano Ronaldo by analyzing tracking data, though camera switching in broadcasts presents a technical hurdle.
- Hot Hand Fallacy: A re-analysis of basketball free-throw data confirmed that players are better at their second free throw, not due to the "hot hand" psychological phenomenon, but due to minor warm-up practice.
- Educational Advice: The founders advise young people to pursue deep curiosity in small topics, allowing that interest to expand naturally into larger fields, rather than forcing specialization or chasing status.
- Interdisciplinary Value: Physics is viewed as a "hard" discipline that builds resilience; students with physics backgrounds often find it easier to adapt to other fields like finance or computer science because they have already learned to handle difficulty.
- Role Models: The conversation highlights the lack of modern public scientific role models comparable to Feynman or Sagan, noting that society needs figures who embody both expertise and approachable enthusiasm.
- Mathematics in Media: The founders emphasize the beauty of mathematics in non-scientific contexts, such as proving the radius of Middle-Earth or analyzing the "Battle of Winterfell" tactics, to demonstrate that math is intellectually democratic and accessible.
- Twitter as Education: Fermat's Library uses Twitter to deliver "dopamine hits" of learning through short, visual factoids, aiming to leverage the platform's high engagement to spark deeper rabbit holes in science.
- Persistence in Stuckness: Both startup building and scientific research require the ability to endure "stuckness," a non-linear process where effort does not immediately yield rewards, which is a common cause of failure for those untrained to handle it.
- Pivoting Strategy: For startups, the founders suggest that when a team runs out of ideas and options, they should pivot rather than continuing to push a failing approach.
- Monetization of Science: The discussion acknowledges the difficulty of monetizing open science without introducing paywalls, suggesting ad-supported models (like Stack Overflow) or public funding as viable alternatives to the current journal industry.
- Data Privacy: The founders express support for transparency in data usage, distinguishing between the ethical collection of data for ads and the lack of user control inherent in platforms like Facebook.
- Future of Collaboration: The "Polymath projects" are cited as proof that large-scale online mathematical collaboration is possible, specifically for problems that can be broken down into incremental steps.