En 30 secondes
Howard argues that research should address practical needs and recounts a transfer-learning project.
- 1IncentivesHoward criticizes publication incentives.
- 2TransferHe values using less data.
- 3Active learningHe describes selecting examples for labeling.
- 4Career pressureHe acknowledges junior researchers’ constraints.
À retenir
- He questions publication incentives.
- Transfer and active learning can reduce resource demands.
- Career pressures affect research choices.
Lire le texte remanié, chapitre par chapitre
Research incentives
Lex Fridman asks fast.ai co-founder Jeremy Howard about the gap between deep-learning theory and practical work.
Lex Fridman asks fast.ai co-founder Jeremy Howard about the gap between deep-learning theory and practical work.
Howard criticizes research incentives that he believes favor incremental work on familiar topics over practical impact. This is his assessment of the field.
Transfer and active learning
He argues that transfer learning can enable useful work with fewer resources and data, and says he considers it insufficiently studied.
He argues that transfer learning can enable useful work with fewer resources and data, and says he considers it insufficiently studied.
Howard also highlights active learning, where a system helps select which examples people should label.
He describes a practical reason for prioritizing difficult examples: labeling everything can be slow and expensive.
A practical project
Howard recounts work with Sebastian Ruder on transfer learning for natural-language processing and says the project began as a practical teaching need.
Howard recounts work with Sebastian Ruder on transfer learning for natural-language processing and says the project began as a practical teaching need.
He says their prototype performed strongly on a benchmark and was later written up and published.
Howard acknowledges that junior researchers may face different incentives and pressure to select safer, familiar topics.
Sur la même chaîne
3:34:27EN
Lex Fridman· 26 oct. 2026Podcast · 4 voixAI, Mathematics, and the Meaning of a ProofInvité : Steven Brunton, Ankur Moitra, Alex Kontorovich · Anime : Lex FridmanLex Fridman and three mathematicians discuss recent AI-generated research, hard open problems and what human understanding contributes.
3:36:56ENLex Fridman· 16 sept. 2026PodcastPsychiatry Has Never Found Its PenicillinAnime : Lex FridmanHistorian Andrew Scull tells Lex Fridman why the field can ease symptoms, has never cured a major mental illness, and keeps paying for overconfidence with patients' lives.
5:15:51ENLex Fridman· 26 août 2026Podcast · 2 voixDHH Did Not Write a Line of Omarchy QuattroInvité : DHH · Anime : Lex FridmanThirteen months after a skeptical conversation with Lex Fridman, the Rails creator describes how Opus 4.5, a swarm of coding agents, and a one-minute Linux installer ended his era of hand-chiseled Ruby.
3:12:58ENLex Fridman· 12 août 2026Podcast · 2 voixOne Cold Shower, Twenty Champions: How Khabib Nurmagomedov Was ForgedInvité : Khabib Nurmagomedov · Anime : Lex FridmanIn a Makhachkala school gym, seventy men shared one shower, and the water was only cold.
3:46:18ENLex Fridman· 28 juil. 2026Podcast · 2 voixIf You Take Slavery Out of the Picture, There Is No Civil WarInvité : Gary Gallagher · Anime : Lex FridmanIn 1860, enslaved people were worth more than every railroad and factory in the United States combined.
3:51:47ENLex Fridman· 30 juin 2026Podcast · 2 voixForty-Six Percent of Emperors Were Overthrown. The Empire Lasted AnywayInvité : Anthony Kaldellis · Anime : Lex FridmanHistorian Anthony Kaldellis walks Lex Fridman through 2,200 years of Roman history and argues that what we call Byzantium was Rome itself.