LA CONVERSATION ORIGINALE

AI researchers debate how close we are to recursive self-improvement

Dwarkesh Podcast · 11 sept. 2026 · 1:37:01

John Schulman discusses uncertainty about generalization, improving experimental methods, and the role of realistic prompt distributions in model distillation.

John Schulman

Ouvrir Audio de l’éditeur

Dwarkesh Podcast

En un coup d’œil

Generalization is difficult to predict[1]Lire le moment probant
Research methods may have substantial room to improve[2]Lire le moment probant
Distillation depends on a realistic prompt distribution[3]Lire le moment probant

Moments clés3

Extraits courts et attribués, avec le contexte permettant de les vérifier. La conversation complète reste chez son éditeur.

9:42

AI research

Generalization is difficult to predict

Extrait original

The whole field relies a lot on generalization and it’s very hard to predict when you’re going to get generalization, or when you’re going to get some kind of out-of-distribution generalization

John Schulman·11 sept. 2026

Cet extrait ne contient pas de contexte supplémentaire. Consultez la conversation originale.

11:34

AI research

Research methods may have substantial room to improve

Extrait original

If you think hard enough, you probably could have expected some of these things beforehand. There is probably some very clever way to do a small-scale experiment that’ll let you build the theory that then will generalize to the large-scale experiment. So I would expect that we’re nowhere near the ceiling of how well you can do research

John Schulman·11 sept. 2026

Cet extrait ne contient pas de contexte supplémentaire. Consultez la conversation originale.

26:48

AI research

Distillation depends on a realistic prompt distribution

Extrait original

If you have a really good realistic prompt distribution for distillation, you can match the big model really well. But if you only have this distribution of easily verifiable tasks, then you can match the big model on all the benchmarks, but you do worse on this broader distribution

John Schulman·11 sept. 2026

Cet extrait ne contient pas de contexte supplémentaire. Consultez la conversation originale.

Source et méthodologie

Ces points de vue renvoient à leurs sources originales. Les reformulations sont signalées et ne sont pas des citations mot à mot.

Ouvrir la transcription ou les documents sources Signaler un problème