AI summary by TwinMind

5. What Do Tom Sawyer and the Founder of Duolingo Have in Common? | No Stupid Questions

Freakonomics Radio Network

29m 21s

Guest Introduction and Background

  • Luis von Ahn: CEO of Duolingo, creator of CAPTCHA
  • Born/raised in Guatemala, came to US for college
  • Former computer science professor at Carnegie Mellon University
  • Created distorted character tests (CAPTCHA) to prove human vs computer identity
  • CAPTCHA stands for "Completely Automated Public Turing Test to Tell Computers and Humans Apart"

CAPTCHA Development and Problem-Solving

  • Origin story: Yahoo chief scientist Udi Manber presented 10 unsolved problems at Carnegie Mellon
  • Core problem: Preventing bots from creating millions of free email accounts for spam
    • Each account limited to 500 messages/day
    • Spammers needed millions of accounts to send 50 million emails
  • Solution approach: Create test computers can administer/grade but not pass
  • Alternative ideas tested: Picture identification (flowers) - too open-ended, required spelling skills
  • Final solution: Distorted text recognition - humans good at it, computers weren't
  • Implementation: 200 million CAPTCHAs typed daily across internet

ReCAPTCHA and Book Digitization

  • Insight: Leverage CAPTCHA typing for useful work
  • Partnership with Google for book digitization project
  • Process: Digital photos of book pages → computer OCR → 30% of words unrecognizable
  • Solution: Send unrecognizable words to people typing CAPTCHAs
  • Users unknowingly helped digitize books while accessing websites
  • Partnerships: New York Times archive digitization
  • Transparency: Question mark button explained the process (most users didn't click)
  • Google acquired the company

Modern CAPTCHA Evolution

  • Current CAPTCHAs use pictures (street signs, stop signs) instead of distorted text
  • Computers improved at reading distorted text over time
  • New purpose: Training image recognition for Google Maps
  • Users help identify stop signs, traffic lights for mapping systems
  • Computer sends uncertain image recognition tasks to humans via CAPTCHA

Duolingo Business Model and Mission

  • Most popular language learning platform globally
  • More US users than entire US public school language programs
  • 10x more people learning Irish on Duolingo than native Irish speakers
  • Fictional languages: High Valyrian (Game of Thrones), Klingon (Star Trek)
  • More High Valyrian learners than Irish learners
  • Private company, venture-funded, $1.5 billion valuation
  • Mission: Free language education (especially English for economic opportunity)
  • English learning can double income potential in non-English countries

Revenue Model

  • 97% of users use free version with ads
  • 3% pay for ad-free subscription
  • 85% of revenue ($90M in previous year, projected to double) comes from 3% paying users
  • Similar model to Spotify (free users supported by premium subscribers)
  • Considering going public

Early Duolingo Translation Model

  • Original concept: Users translate real content while learning
  • Contracts with CNN and BuzzFeed for news translation
  • English articles given to Spanish speakers learning English
  • Multiple users translate same paragraphs, computer combines translations
  • Problem: Translation industry has declining prices, improving computer translation
  • Model abandoned due to poor business economics

Academic Publishing and Research Volume

  • 1-2 million scientific articles published annually (fact-check: actually ~3 million)
  • Typical article receives 0-2 citations
  • Problem: Scientists chase metrics (paper count, citations) rather than impact
  • Historical context: 20-30 years ago, tenure required small number of quality papers
  • Current system: Long CVs with many articles required for job interviews
  • Google Scholar made citation counting easier, became worth metric

Academic Quality vs Quantity Debate

  • Example: Manuel Blum (von Ahn's PhD advisor, Turing Award winner)
  • Published couple papers per year, each highly impactful
  • Focused on deep understanding rather than publication volume
  • Teaching method: Claimed not to understand until student explained clearly
  • Current focus: Understanding consciousness
  • Contrast with researchers publishing 14+ papers per year at single conference

Computer Science Field Evolution

  • Early computer scientists more philosophical, focused on fundamental questions
  • Modern field more pragmatic due to high-paying industry jobs (Google, Facebook)
  • Shift from "what does it mean to be human" to vocational training
  • Still some researchers pursuing deep questions, but minority

Solutions for Academic Publishing Problems

  • Need incentive system rewarding quality over quantity
  • Suggestion: "Relegation" system - tenure removal for too many low-citation papers
  • Translation problem: Most academic writing incomprehensible to lay people
  • Journal costs: $20,000+/year (fact-check: can be $10M+ for major publishers)
  • Some progress: NBR weekly digests, open access requirements
  • Academic-to-public translation role important (like Angela Duckworth's work)

Creativity and Volume Theory

  • Dean Keith Simonton's "equal odds rule"
  • Linear relationship: Want 1 good idea, generate 10 ideas
  • 99% of creative output may be "crap" but volume necessary for breakthrough
  • Tension between quality focus and creative volume needs

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