John Schulman quote, Sep 11, 2026

Chief Scientist, Thinking Machines Lab · Sep 11, 2026 · Interview, Dwarkesh Podcast: AI researchers debate how close we are to recursive self-improvement

In fact, I remember in the early OpenAI days having the intuition that just minimizing log loss wasn’t going to get you to intelligence.

Primary source

Words found on the source page on Sep 25, 2026.

More from John Schulman

Date	Quote	Context	Source
Sep 11, 2026	Basically anything that can be learned through RL can be distilled very easily, because it’s a small number of bits.Sep 11, 2026· Interview	InterviewDwarkesh Podcast: AI researchers debate how close we are to recursive self-improvement	Source
Sep 11, 2026	Right now, you don’t get explosive growth in capabilities because you still get bottlenecked enough when you’re trying to do research and engineering.Sep 11, 2026· Interview	InterviewDwarkesh Podcast: AI researchers debate how close we are to recursive self-improvement	Source
Sep 11, 2026	So I would expect that we’re nowhere near the ceiling of how well you can do research.Sep 11, 2026· Interview	InterviewDwarkesh Podcast: AI researchers debate how close we are to recursive self-improvement	Source
May 15, 2024	Even in one or two years, we'll find that the models can do a lot more involved tasks than they can do now.May 15, 2024· Interview	InterviewDwarkesh Podcast: Reasoning, RLHF, and Plan for 2027 AGI	Source
May 15, 2024	Given the relatively small number of entities who could train the largest models, it does seem possible to coordinate.May 15, 2024· Interview	InterviewDwarkesh Podcast: Reasoning, RLHF, and Plan for 2027 AGI	Source
May 15, 2024	If AGI came way sooner than expected we would definitely want to be careful about it.May 15, 2024· Interview	InterviewDwarkesh Podcast: Reasoning, RLHF, and Plan for 2027 AGI	Source

Date	Quote	Context	Source
Sep 24, 2026	An AI with more real layers can actually be “smarter” in the conventional sense of the term.Scott AlexanderSep 24, 2026· Post	PostThe Specter Of Neuralese	Source
Sep 24, 2026	For 100 steps in a row, the AI can shoot delicate subtle ideas from layer to layer at the speed of light. Then there’s one step where it has to encode them into twenty-six glyphs invented by Phoenician turquoise miners in 1800 BC. Then it has to re-encode the Phoenician glyphs into delicate subtle lightspeed ideas before it can do anything else.Scott AlexanderSep 24, 2026· Post	PostThe Specter Of Neuralese	Source
Sep 24, 2026	The science of reading these thoughts is a subfield of AI interpretability, which is still in its infancy.Scott AlexanderSep 24, 2026· Post	PostThe Specter Of Neuralese	Source
Sep 23, 2026	This trains the AIs to be focused on task success, which naturally risks including things like reward-hacking, cheating, and single-minded pursuit of stated goals at the expense of ethical injunctions.Scott AlexanderSep 23, 2026· Post	PostMysteries Of AI Generalization	Source
Sep 19, 2026	It is important to not confuse massive steps in inference-time scaling, a dynamic which should be fairly predictable, with being the outputs of RSI, which is highly uncertain.Nathan LambertSep 19, 2026· Post	PostWhy I still haven't bought into true RSI	Source
Sep 17, 2026	If you have five hours, you’re probably going to do a lot better. The AI models are pretty similar.Noam BrownSep 17, 2026· Interview	InterviewDwarkesh Podcast: Agent swarms, alignment, and recursive self-improvement	Source

About this quote

Said by: John Schulman, Chief Scientist, Thinking Machines Lab
Where: Dwarkesh Podcast: AI researchers debate how close we are to recursive self-improvement
Topics: Research
Company: Thinking Machines Lab
Source: dwarkesh.com/p/john-beren-charlie

A quote is one moment, not the whole argument. Read it in context at the source.

Sources: the primary source linked on each quote, checked weekly. A quote is one moment; read the source. Logos via logo.dev; trademarks belong to their owners.

More from John Schulman

On the same topics

New quotes by email