scripod.com
Ilya Sutskever – We're moving from the age of scaling to the age of research

Overview

Shownote

Highlights

Transcript

Chapters

Pins

Ilya Sutskever – We're moving from the age of scaling to the age of research

Dwarkesh Podcast

5 DAYS AGO
Dwarkesh Podcast

Dwarkesh Podcast

5 DAYS AGO
OverviewShownoteHighlightsTranscriptChaptersPins

Shownote

Ilya & I discuss SSI’s strategy, the problems with pre-training, how to improve the generalization of AI models, and how to ensure AGI goes well. Watch on YouTube; read the transcript. Sponsors * Gemini 3 is the first model I’ve used that can find conne...

Highlights

In this deep-dive conversation, Ilya Sutskever explores the frontiers of AI research, focusing on the limitations of current models and the path toward safe, general, and ultimately superintelligent systems. The discussion moves beyond benchmark performance to examine the foundational challenges in learning efficiency, generalization, and alignment.
16:34
Emotions modulate the human value function in an evolutionarily hard-coded way
18:54
Gemini 3 may have discovered more efficient pre-training methods beyond simple scaling.
31:26
Gemini 3 modeled RL information gain as entropy and generated testable code
42:39
Deploying AI systems helps improve safety through real-world feedback and failure correction.
46:55
A safe superintelligence should learn continuously through deployment, not be statically released.
1:02:03
Building AI that cares about sentient life is more important than maintaining human control.
1:18:15
Strategies will converge as AI becomes more powerful; all should aim for an aligned, sentient-caring superintelligent AI.
1:30:35
Self-play is a special case of agent competition that can incentivize diverse approaches

Chapters

Explaining model jaggedness
00:00
Emotions and value functions
09:39
What are we scaling?
18:49
Why humans generalize better than models
25:13
SSI’s plan to straight-shot superintelligence
35:45
SSI’s model will learn from deployment
46:47
How to think about powerful AGIs
55:07
“We are squarely an age of research company”
1:18:13
Self-play and multi-agent
1:30:26
Research taste
1:32:42

Transcript

Ilya Sutskever: you know what's crazy, that all of this is real? Yeah, don't you think so? Like all this ai stuff and all this big era? Yeah, that it's happened, like, isn't it straight out of science fiction? Dwarkesh Patel: Yeah, another thing that's cr...