Lex Fridman PodcastLuís and João Batalha: Fermat's Library and the Art of Studying Papers | Lex Fridman Podcast #209
CHAPTERS
- 0:00 – 2:00
Open access to science: why paywalls undermine discovery
Lex opens with a case for making scientific papers freely accessible, arguing that paywalls slow collaboration and distort incentives. He frames journals as gatekeepers that monetize publicly funded research while relying on unpaid peer review labor.
- •Most research remains behind paywalls despite being publicly funded
- •Journals monetize access while peer review is largely unpaid labor
- •Free dissemination accelerates collaboration and progress
- •Digital distribution costs are low; alternative funding models could work
- 2:00 – 5:28
Backstories as a missing “section” of papers: from Feynman to Goodfellow
The conversation begins with the idea that papers should include a “backstory” section to capture the human path to discovery. Luís shares examples where small moments, chance interactions, or personal struggles shaped major breakthroughs.
- •Backstory adds context and corrects the myth of isolated ‘eureka’ papers
- •Feynman’s plate-wobble insight leading toward QED
- •Ian Goodfellow’s GAN idea emerging in a bar; value of AMAs with authors
- •Cédric Villani’s account of how incremental interactions yield breakthroughs
- 5:28 – 10:25
Insight vs grind: field maturity, memorability, and romanticizing discovery
Lex and João discuss whether breakthroughs are mostly brilliant flashes or long-term grind, and how this varies by field maturity. They explore how stories (true or not) help cement knowledge and make algorithms and ideas easier to remember.
- •Low-hanging fruit is more common in fast-growing fields like ML
- •Field maturity reduces frequency of sudden ‘bar-night’ breakthroughs
- •Dijkstra’s shortest path origin story and how it aids recall
- •Backstory can be a learning tool—making ideas ‘stick’
- 10:25 – 17:14
Great scientists as people: Newton’s alchemy, Dyson’s humility, curiosity as fuel
The discussion turns to scientific personalities and how their quirks relate to creativity. Newton’s alchemy and biblical interpretation, and Freeman Dyson’s humility and curiosity, highlight how ‘non-scientific’ interests and mindset shape scientific work.
- •Newton’s alchemy and theology as part of the historical context of science
- •Alchemy as ‘dreaming the impossible’—and why it matters
- •Freeman Dyson’s humility and desire to open questions vs answer them
- •Curiosity persisting into old age as a hallmark of great thinkers
- 17:14 – 23:57
What Fermat’s Library is: annotating papers to unlock the margins
Lex asks for the mission and product: Fermat’s Library is presented as a platform to annotate papers with rich, contextual explanations. The brothers describe how it grew out of a journal-club habit and a desire to make dense research more accessible.
- •Annotations in the margins: LaTeX/Markdown sidebars tied to exact passages
- •Papers are dense and constrained; annotations can evolve clarity over time
- •Origin in a small journal club across different backgrounds (CS/physics)
- •Broader mission: help move science ‘in the right direction’ by improving access
- 23:57 – 25:56
Products around the core: Journal Club, Margins, and the Librarian arXiv extension
They break down Fermat’s Library’s ecosystem: weekly annotated papers, free tools for others to run clubs, and a browser extension that overlays features on arXiv. Lex emphasizes how much better the experience becomes compared to arXiv’s minimal interface.
- •Journal Club: weekly annotated paper releases across fields
- •Margins: the underlying annotation software, offered free for others to use
- •Librarian: an arXiv overlay enabling comments, references, and BibTeX extraction
- •Why adoption in science tools can be slow despite obvious usability benefits
- 25:56 – 37:14
arXiv, preprints, and the missing collaboration layer (plus bioRxiv history)
They discuss why preprint repositories matter—especially given multi-year journal delays in fields like math—and why arXiv needs richer layers for collaboration and quality signals. Luís recounts the ‘Information Exchange Group’ in 1960s biology and how journals suppressed early preprints.
- •Preprints unlock access during long journal publication cycles
- •Perelman’s Poincaré work as a landmark example of arXiv-first impact
- •Need for tooling on top of arXiv: comments, evaluation signals, code/data links
- •1960s biology preprint experiment was pressured into cancellation; long-term impact on adoption
- 37:14 – 45:17
Scientific publishing critique: peer review, impact factor, and ‘gaming the metrics’
The conversation becomes a systemic critique: journals collect rents while governments fund research, review, and buy access back. Luís argues impact factor turns papers into accounting units, creating a ‘tyranny of metrics’ that incentivizes clickbait and misconduct.
- •Bizarre circular economy: public funds research, review, and subscriptions
- •Impact factor as ‘nobility title’—papers are ‘born with impact’ before reading
- •Counting citations creates a market; parallels to PageRank and SEO
- •Two ‘economies’: low-impact volume pressure vs luxury-journal media attention incentives
- 45:17 – 57:18
Toward crowdsourced review and new incentives: Wikipedia, Stack Overflow, Polymath
Lex explores whether science can adopt internet-scale review and contribution models like Amazon reviews, Wikipedia, or Stack Overflow. Luís points to Polymath as evidence that massively collaborative research can work for problems that decompose into steps, while acknowledging incentive conflicts in academia.
- •Crowd review could improve robustness vs small anonymous reviewer sets
- •Power users and gamification can drive knowledge platforms
- •Polymath projects show online collaboration works for decomposable math problems
- •Academic career scarcity and competition drive metric gaming and conservatism
- 57:18 – 1:00:55
Paper 2.0: beyond static PDFs—code, data, video, and even Twitter threads
They imagine how papers might evolve: richer formats integrating code, datasets, and explainer media. They discuss experiments with publishing ideas as tweet threads, while also acknowledging why the constraints and formalism of papers still matter.
- •Modern research outputs exceed what PDF/LaTeX can express well
- •Potential ‘paper 2.0’ interface: interactive, multimedia, reproducible artifacts
- •Experiments with short Twitter-thread ‘publications’
- •Constraints of papers can be beneficial for clarity and concreteness
- 1:00:55 – 1:07:06
How to read a paper: conclusions-first, reference hopping, and trusting the chaos
The brothers and Lex compare strategies for reading difficult papers, emphasizing that papers are not optimized for comprehension. They discuss approaches inspired by Feynman and the practical reality of jumping through citations until a ‘quantum jump’ in understanding occurs.
- •Difficulty isn’t always about the material—often it’s the paper format
- •Feynman-style: start from the conclusion and reconstruct the path yourself
- •Expect to open many tabs and traverse references; breakthroughs come suddenly
- •Related work sections provide context and reveal field tensions and agendas
- 1:07:06 – 1:15:27
Taking notes and teaching your future self: annotations, audio, and verbalizing
They discuss note-taking as a tool for memory and re-learning, with a focus on writing for your future self right after understanding something. Lex adds that audio notes can ‘reload’ the mental context better than text, and João emphasizes that verbalizing forces structure.
- •Underestimate how much you’ll forget; capture understanding immediately
- •You are the best teacher for your future self right after a breakthrough
- •Audio/video can preserve context and nuance that text notes lose
- •Explaining aloud forces structure and clarifies what you actually know
- 1:15:27 – 1:36:13
Favorite Fermat’s Library papers: ancient tunnels, Fermi’s blast math, Bitcoin fundamentals
They share standout papers and what annotations add: an ancient Greek tunnel construction mystery, Fermi’s one-page Trinity test energy estimate, and the Bitcoin whitepaper as Fermat’s first annotated paper. The theme is making dense or context-heavy ideas accessible by embedding the missing prerequisites in the margins.
- •Samos tunnel paper: correcting 2,000-year-old misconceptions; plausible surveying methods
- •What makes a ‘good annotation’: historical context, instruments, visualization, updates from later research
- •Fermi’s Trinity report: paper scraps to estimate blast energy; applying the method to Beirut explosion video
- •Bitcoin paper annotations: public key crypto, hash functions, Merkle trees, and community-added layers over time
- 1:36:13 – 1:56:03
Math culture and scientific ideals: Perelman’s protest, credit drama, and Tao’s openness
The conversation turns to the sociology of mathematics via Perelman’s arXiv publishing, rejection of prizes, and distaste for status games. They then highlight Terence Tao as a modern role model for open science and honest discussion of struggle, including a blog-comment-driven path to solving the Erdős discrepancy problem.
- •Perelman publishing on arXiv as protest against publishing gatekeeping
- •Fields Medal and Clay Prize rejection tied to attribution drama and principles
- •Russian math culture: doing math for intellectual pleasure, not money
- •Terence Tao’s blog as open collaboration platform; vulnerability about learning and struggle
- 1:56:03 – 2:05:51
Fermat’s Library on Twitter: micro-learning, dopamine, and ‘math drug dealer’ dynamics
Lex praises Fermat’s Library’s Twitter feed for delivering compact ‘eureka’ moments through visuals, facts, and puzzles. Luís explains the philosophy: math has an ‘infinite half-life,’ Twitter is underused for learning, and the challenge is making content that is easy to consume but hard to craft.
- •Math validity persists for centuries (Euclid) unlike many fields
- •Twitter as a learning platform: fast, addictive, zero ‘dropout’ vs courses
- •Content design aims to trigger tiny eureka moments and rabbit holes
- •Balancing rigor, visual appeal, and breadth across math/physics/history
- 2:05:51 – 2:54:52
Startup lessons from science: non-linear rewards, pivoting, and long-horizon building
They map research psychology onto startups: the default state is being stuck, rewards are non-linear, and perseverance matters. João argues Fermat’s Library benefits from a 20-year mindset and slower growth, since infrastructure for science can’t be forced into typical 3–5 year startup timelines.
- •Startups resemble research: many dead ends before a breakthrough
- •Non-linear reward functions are demoralizing; schooling trains the opposite
- •Pivot vs persist is hard; one heuristic is ‘pivot when you run out of ideas’
- •Fermat’s Library as a long-horizon project; some ecosystems (like arXiv) take decades
