AI summary by TwinMind

Luís and João Batalha: Fermat's Library and the Art of Studying Papers | Lex Fridman Podcast #209

Lex Fridman

2h 54m 52s

Introduction and Background

  • Conversation with Luis and Joao Batala, co-founders of Fermat's Library
  • Fermat's Library: platform for annotating scientific papers
    • Named after Fermat's Last Theorem - famous margin note that didn't fit
    • Allows professional scientists, academics, citizen scientists to annotate equations, figures, ideas
  • Discussion of scientific publishing paywall problem
    • Over 70% of published research articles behind paywall
    • Journals create paywalls using unpaid peer review labor
    • Cost should be minimal in digital age, covered by donations/ads/public funding

Backstory Importance in Scientific Papers

  • Luis suggested papers should have "backstory sections" like abstracts
  • Human context crucial for understanding scientific ideas
  • Feynman QED paper example:
    • Feynman went through dark phase, lost enthusiasm for physics
    • Saw plate thrown in Cornell cafeteria - noticed wobbling and Cornell symbol rotation
    • Led to equations of motion work → electron orbits in relativity → QED paper → Nobel Prize
  • Ian Goodfellow GAN paper: idea conceived in bar, implemented in single night
  • Cedric Villani book: detailed path from PhD to Fields Medal, importance of random interactions

Low-Hanging Fruit in Science

  • Discussion of breakthrough moments vs. grinding work
  • Machine learning field growing rapidly: 50% of ML papers on arXiv published in last 12 months
  • Field maturity affects frequency of eureka moments
  • Dijkstra's shortest path algorithm: developed in 20 minutes at Amsterdam coffee shop without pen/paper
  • Simplicity constraint forced elegant solution

Inspiring Scientists and Personalities

  • Freeman Dyson:
    • Extremely humble despite major accomplishments
    • Email exchanges showed curiosity even at 90+ years old
    • Created prime number portrait for his birthday using photo converted to prime
  • Richard Feynman: exceptional scientist and teacher, socially adept
  • Childlike curiosity as common trait among great scientists
  • Daniel Kahneman example of maintained curiosity in older age

Fermat's Library Platform Details

  • Core components:
    • Journal Club: weekly annotated papers across physics, CS, math
    • Margins: software for university journal clubs
    • Librarian: browser extension overlay for arXiv with commenting interface
  • Interface exceptionally well-designed for paper annotation
  • Supports LaTeX and Markdown in annotations
  • Papers become more accessible over time through community annotations
  • Challenge: people afraid to add "imperfect" annotations to well-annotated papers

Scientific Publishing System Problems

  • Traditional publishing vs. scientific publishing comparison:
    • Normal: journals pay authors, reviewers, distributors
    • Scientific: government funds research → scientists give work free → other scientists review free → government buys back through university libraries
  • Impact factor creates "tyranny of metrics"
    • Papers born with impact before anyone reads them
    • Goal becomes publication, not long-term impact
  • Two economies in publishing:
    • Luxury journals (Nature ~40 impact factor) need media attention
    • Low-impact journals for quantity publishing
  • Gaming the system through clickbait titles, inflated claims

Peer Review and Metrics Discussion

  • Current peer review: handful of people, often anonymous, limited incentive for quality
  • Comparison to Amazon review system: thousands of contributors, better representation
  • Crowdsourcing review could allow broader participation beyond narrow expertise
  • Heisenberg uncertainty principle for metrics: as soon as you measure, people game it
  • Need more resilient metrics, but underlying problem is academic job scarcity

Archive and Preprints

  • Archive: 30-year-old repository, linear growth, fundamental to modern science
  • Math papers can take 3 years from submission to publication
  • Machine learning field evolves so fast people work from preprints
  • Biology historically suppressed preprints (1960s NIH experiment killed by journal pressure)
  • Only 2% of biology papers appear as preprints first
  • European Union and Gates Foundation requiring open access

Paper Format Evolution

  • Papers fundamentally unchanged since first Nature issue
  • Modern papers include code, data, videos but still constrained to PDF format
  • Need for Paper 2.0: more dynamic, multimedia format
  • Twitter threads as alternative publication format experimentation
  • Constraints and formalism of papers still valuable for concrete idea expression

Reading Papers Effectively

  • Papers not optimized for understanding due to length/format restrictions
  • Feynman method: read conclusion first, try to derive it independently
  • Fermi approach: back-of-envelope calculations
  • Accept initial low understanding, trust the process through multiple readings
  • Breadth-first approach: conclusion → related work → references
  • Related work sections often most accessible, provide field context
  • Note-taking importance: you're best teacher for your future self after forgetting

Notable Annotated Papers

  • Tunnel of Samos (6th century BC):
    • Mathematical analysis of ancient engineering problem
    • 2000-year-old explanation was wrong
    • Likely used iron sight technique with three sticks
    • No tunnels built for 1000 years after Roman fall
  • Bitcoin paper: first on Fermat's Library
    • Extremely dense, precise language
    • Annotations explain cryptography basics, hash functions, Merkle trees
  • Fermi Trinity bomb calculation:
    • One-page classified report using paper pieces in blast
    • ~20% accuracy using displacement method
    • Applied same technique to Beirut explosion analysis
  • Freeman Dyson Dyson Sphere paper:
    • One-page concept for stellar energy extraction
    • Stability analysis, detection via infrared radiation
    • Complete framework from concept to detection method

Science Communication and Twitter

  • Math content has "infinite half-life" - Euclid still valid after 2000 years
  • Twitter underexplored for learning: dopamine hits from understanding
  • Zero percent dropout rate vs. traditional courses
  • Example: 23456789 largest prime with consecutive increasing digits
  • Variety: numeric facts, physics, history, visual animations
  • Goal: get people "hooked" on mathematics like drug dealers
  • Belief that everyone can love math/physics with proper explanation

Startup Insights

  • Non-linear reward functions: effort doesn't guarantee proportional results
  • Sisyphus boulder analogy: work can yield no reward
  • Most people not equipped for this psychological challenge
  • Framework: pivot when you run out of ideas to try
  • Many broken systems in science publishing create opportunities
  • Long-term vision (20 years) vs. typical 3-5 year startup timeline
  • Some problems require longer development cycles than venture funding allows

Perelman and Mathematical Purity

  • Rejected Fields Medal due to credit attribution disputes
  • Published Poincaré conjecture proof on arXiv, not journals
  • Protest against scientific publishing industry paywalls
  • Platonic ideal of mathematician: cares only about mathematics
  • Russian mathematical tradition: doing math for intellectual pleasure
  • Distinction between professional sports (money-focused) vs. Olympics (pure artistry)
  • Removing money from equation allows focus on excellence

Soccer Analysis and Sports Science

  • Game of Thrones Battle of Winterfell differential equations
  • 1.7 kill ratio required for living army to defeat dead army
  • Mathematical analysis makes fantasy intellectually democratic
  • Sports provide constrained systems for psychological analysis
  • Basketball hot hand phenomenon: no correlation found in free throws
  • Computer vision challenges in soccer: camera perspective switches
  • Potential for crowdsourced sports analysis datasets

Advice for Young People

  • Poincaré quote: teach children to be enthusiastic about world's mysteries
  • Follow curiosity iteratively: numbers → prime factorization → number theory → cryptography → bitcoin
  • Start small, let fire build naturally
  • Don't pursue for status/fame/money - pursue for happiness and meaning
  • Society should allow curiosity fulfillment
  • European vs US education: more rigid vs. more exploratory freedom
  • Physics background provides confidence to tackle other difficult fields
  • Mathematics/physics should be as admired as other skills in society

TwinMind

Get summaries like this for your own meetings

TwinMind takes notes in the background during meetings, lectures, and calls, then writes the summary for you.

Free plan to get started

No bot joins your calls

Phone, desktop & Chrome apps

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome