Young AI Researcher Disrupting the Entire Industry - Ali Behrouz

About this episode

Ali Behrouz is an AI researcher and the author of two of the most talked about papers in machine learning right now, Titans and Nested Learning. He is a PhD student at Cornell University and a researcher at Google, and his work takes direct aim at one of the field's biggest questions: what comes after the transformer.

His paper Nested Learning, which Google's Jeff Dean called a possible paradigm shift, rethinks the whole deep learning stack. In this conversation we get into the back story behind Titans and Nested Learning, how Ali thinks about transformers and where they fall short, the idea that all of deep learning is really associative memory, and his newer work on giving models a kind of "sleep" phase to consolidate what they learn. One of the most insightful conversations we have had on where AI is heading.

Ali was named as the best AI researcher in 2025 by Second Group: Second Prize Awards 2025, Celebrating the Best Iranian AI Scientists, Roboticist, CEOs & Startups

Ali Behrouz : Ali BehrouzGitHubhttps://abehrouz.github.io

Ardalan Javadi: https://www.linkedin.com/in/ardalanjam1369/

Farzam Hejazi: https://www.linkedin.com/in/farzam-hejazi-5b608081/

Subscribe to our Newsletter : https://substack.com/@seondnewsletter

 

00:00 Introduction 

03:26 What is Intelligence? 

05:51 Academic Journey

11:57 From Blockchain to Brain Networks

17:21 The Allure of Computational Social Science

21:02 The Genesis of TITAN

24:39 Human Memory Systems vs. AI Caching

32:36 The Disconnect: Solving Anterograde Amnesia in LLMs

41:40 Mechanics of TITAN

47:37 The Viral Aftermath: Dealing with Hype and Haters

52:09 Why Transformers Dominate & The Rebranding Epidemic

57:24 Scaling Laws and the Limitations of Pure Compression

64:05 Anthropic vs. OpenAI

66:06 Architectural Frontiers: Loop Transformers & Mixture of Depths

69:01 Moving Beyond TITAN

77:51 Nested Learning: Emulating Brain Frequencies

81:44 Memory Distillation and AI Sleep Cycles

83:24 Career Advice: Scientific Progress as "Old Ideas + New Noise"

 

References

https://www.seltzer.com/margo/

https://en.wikipedia.org/wiki/Large-scale_brain_network

https://www.geeksforgeeks.org/machine-learning/introduction-to-recurrent-neural-network/

https://en.wikipedia.org/wiki/Mamba_(deep_learning_architecture)

 

About Us : 

Linkedin : https://www.linkedin.com/company/secondpodcast/

Twitter : /thesecondpod   

Instagram : https://www.instagram.com/thesecondpod/

#Secondpod #Second_Pod #پادکست ـ دوم #secondpod #پادکست