Latest Interviews

Dileep George: Brain-Inspired AI | Lex Fridman Podcast #115

Dileep George, Lex Fridman

Vicarious and its researchers advocate for a recursive cortical network architecture that prioritizes biological plausibility through feedback loops and local learning rules rather than the massive scaling and backpropagation of current deep learning models. This approach has enabled the system to achieve state-of-the-art performance on CAPTCHAs and few-shot learning tasks by modeling the causal structure of the world instead of memorizing statistical correlations in text. By integrating perception, cognition, and language into a unified generative framework, the work challenges the limitations of feed-forward transformers and proposes a path toward true artificial general intelligence grounded in physical and temporal reasoning.

Russ Tedrake: Underactuated Robotics, Control, Dynamics and Touch | Lex Fridman Podcast #114

Russ Tedrake, Lex Fridman

MIT roboticist and Toyota Research Institute Vice President Russ Tedrick advocates for leveraging passive dynamics and underactuated systems to enhance robotic efficiency and safety in complex environments. His work bridges theoretical rigor with practical application through the development of the Drake simulation framework and lessons learned from the DARPA Robotics Challenge, where failures highlighted the critical need for robust state estimation and diverse testing regimens. Looking forward, Tedrick predicts the rapid adoption of soft, fleet-learning robots in home and elder care settings, driven by a philosophy that prioritizes fundamental physics over purely data-driven approaches.

Manolis Kellis: Human Genome and Evolutionary Dynamics | Lex Fridman Podcast #113

Manolis Kellis, Lex Fridman

Genomics expert Manolis Kellis explores fundamental biological principles, characterizing life as a robust, fault-tolerant digital system while utilizing evolutionary analysis to determine the natural origin and adaptive dynamics of SARS-CoV-2. He contrasts biological complexity with engineering paradigms to advocate for a future where education prioritizes epistemology over rote knowledge and where brain-computer interfaces rely on human neural plasticity to adapt to machine languages. The discussion further examines the philosophical meaning of life, positing that the continuous quest for connection and utility, combined with the biological completion found in parenting, serves as the primary driver of human existence.

Ian Hutchinson: Nuclear Fusion, Plasma Physics, and Religion | Lex Fridman Podcast #112

Ian Hutchinson, Lex Fridman

A comprehensive presentation explores the scientific fundamentals of nuclear and plasma physics while highlighting the International Thermonuclear Experimental Reactor (ITER) as the first project capable of sustaining a controlled burning plasma in the near future. Beyond technical barriers like magnetic confinement and the comparison between fission and fusion energy, the speaker critically addresses existential threats such as overpopulation and challenges the notion that technology alone can resolve sociological or moral crises. The discussion further integrates theological perspectives, arguing against scientism and transhumanism while defining the purpose of human life through relational fidelity to God and community rather than indefinite technological extension.

The most controversial Python feature | Walrus operator

Guido van Rossum

Introduced in Python 3.8 via PEP 572, the Walrus operator (`:=`) functions as a named assignment expression that condenses variable assignment and conditional checks into single lines for applications like regular expressions and list comprehensions. Its contentious adoption fueled a deep community divide over syntax and readability, directly contributing to Guido van Rossum's resignation as Benevolent Dictator for Life. Although the feature challenges traditional Zen of Python principles, it remains a technically elegant tool for code condensation in data science when applied correctly.

Richard Karp: Algorithms and Computational Complexity | Lex Fridman Podcast #111

Richard Karp, Lex Fridman

Renowned theoretical computer scientist Richard Karp, recipient of the 1985 Turing Award, discusses his foundational contributions to NP-completeness theory and algorithms for string matching and stable matching while critiquing the limitations of current AI capabilities and the complexity of biological data. He addresses the unresolved P versus NP conjecture, arguing that efficient solutions likely do not exist for combinatorial problems despite their practical tractability in many real-world scenarios. The conversation further explores the ethical implications of genetic engineering, the necessity of rigorous preparation in teaching, and the distinct gap between empirical machine learning success and formal algorithmic guarantees.

Jitendra Malik: Computer Vision | Lex Fridman Podcast #110

Jitendra Malik, Lex Fridman

Renowned Berkeley professor Jitendra Malik argues that computer vision has historically underestimated human cognitive complexity, noting that deep learning's "tabula rasa" approach fails to capture the evolved, action-guided nature of biological sight. He warns that near-term autonomous driving remains unlikely due to the critical need for sophisticated reasoning to handle rare edge cases, a gap that static image analysis cannot bridge without active, exploratory learning mechanisms. Malik advocates for shifting the field toward multi-task architectures and video understanding to solve these fundamental "Hilbert problems," while cautioning that the immediate societal threat lies in the deployment of unsafe, biased systems rather than the distant arrival of superhuman artificial general intelligence.

C Programming Language | Brian Kernighan and Lex Fridman

Brian Kernighan, Lex Fridman

Brian Kernighan and Dennis Ritchie co-authored "The C Programming Language" in 1977 to capitalize on Unix's rapid expansion by providing the definitive guide to a language designed for efficiency and portability. The book's unique pedagogical approach prioritized practical, real-world Unix utilities like copy and grep over abstract syntax, establishing a standard for teaching programming through immediate, tangible utility. This collaboration successfully created a ubiquitous ecosystem where C applications could run across diverse systems, cementing the language's longevity through a positive feedback loop with the operating system it helped define.

Brian Kernighan: UNIX, C, AWK, AMPL, and Go Programming | Lex Fridman Podcast #109

Brian Kernighan, Lex Fridman, Dennis Ritchie

This discourse explores the historical evolution of Unix from its 1969 Bell Labs origins through the development of Linux and the creation of the C programming language by Ken Thompson and Brian Kernighan. The conversation details the enduring "everything is a file" design philosophy, the collaborative culture that spawned tools like AWK and AMPL, and the transition to modern languages such as Go. Finally, the analysis addresses critical contemporary challenges, including the ethical risks of artificial intelligence, the consequences of hardware scaling limits, and the necessity of broad computer literacy in an increasingly automated digital society.

Sergey Levine: Robotics and Machine Learning | Lex Fridman Podcast #108

Sergey Levine, Lex Fridman

The event examines the persistent intelligence gap in robotics, arguing that true common sense requires physical interaction and off-policy learning rather than mere text processing. It contrasts rigid, modular approaches with modern end-to-end systems that leverage simulation and lifelong learning to handle the unpredictable variables of real-world environments. Furthermore, the discussion reframes reinforcement learning from narrow task optimization to broad cognitive tool acquisition, while addressing safety concerns regarding unintended consequences and the existential risks posed more by human misuse than by autonomous systems themselves.

Peter Singer: Suffering in Humans, Animals, and AI | Lex Fridman Podcast #107

Peter Singer, Lex Fridman

Philosopher Peter Singer draws on his Holocaust-survivor family history to articulate a utilitarian framework that prioritizes the reduction of conscious suffering across humans, animals, and potentially future artificial intelligences. He defines speciesism as a moral bias analogous to racism, arguing that the capacity to suffer is the sole criterion for moral consideration while advocating for effective altruism strategies to address immediate global crises. Although he remains skeptical of imminent existential risks from superintelligence, Singer emphasizes the ethical imperative to donate a portion of income and pursue high-impact careers to maximize positive utility in a finite world.

Cognition Is a Function of the Environment | Matt Botvinick and Lex Fridman

Matt Botvinick, Lex Fridman

The presentation examines the computational mechanisms underlying human cognition and artificial intelligence, contrasting the Turing machine's universal emulation capabilities with the brain's adaptive, capacity-limited nature. It highlights how collective intelligence emerges through distributed convergence while debating whether individual or community structures best define intelligence. Furthermore, the discussion emphasizes that cognitive development relies heavily on the dynamic interaction between neural architectures and environmental structures, as evidenced by DeepMind's self-play research which identifies competitive agent interactions as essential for complex learning.

Matt Botvinick: Neuroscience, Psychology, and AI at DeepMind | Lex Fridman Podcast #106

Matt Botvinick, Lex Fridman

Neuroscientist Matthew Botvinick outlines a unified framework bridging the gap between high-level cognitive psychology and granular neural mechanisms, emphasizing the prefrontal cortex's role in meta-learning and distributed dopamine coding for behavioral flexibility. He argues that advancing artificial intelligence requires shifting focus from mere competence to engineering "warmth" and human-like adaptability, ensuring value alignment through deep social understanding. This synthesis aims to redefine the future of AI safety, moving beyond risk mitigation to actively designing systems capable of complex, positive human interaction and enlightenment.

I'm back at it: 1,000 total push-ups, pull-ups, squats every day

After recovering from an upper-body injury sustained during the eighth day of a 30-day challenge, the creator resumed a rigorous regimen of 34 daily rounds comprising push-ups, pull-ups, and squats to test mental endurance over conventional health metrics. Rejecting standard wellness advice in favor of high-risk self-discovery, the speaker frames this physical strain as a method to develop a trainable "mental muscle" capable of withstanding intense daily grind. While prioritizing core passions in artificial intelligence, the creator advocates for academics and engineers to adopt similar high-volume physical habits to maintain immunity and clear the mind.

You Are Your Own Existence Proof (Karl Friston) | AI Podcast Clips with Lex Fridman

Karl Friston, Lex Fridman

The Free Energy Principle proposes that any system maintaining a boundary against its environment must minimize variational free energy to maximize the evidence of its own existence, a mathematical necessity that applies from non-living oil drops to biological organisms. This framework distinguishes living systems by their possession of a Markov blanket enabling active inference, where agents use internal models to plan actions and resolve uncertainty rather than passively processing data like traditional machine learning models. Ultimately, the theory posits that consciousness and self-awareness emerge as complex generative models required for distinguishing an agent from others in a social world, driving the fulfillment of specific existential narratives to maintain structural integrity.