Published May 15, 2024

John Schulman (OpenAI Cofounder) - Reasoning, RLHF, & Plan for 2027 AGI

John Schulman, cofounder of OpenAI, delves into AI's future as proactive collaborators, discussing the need for regulation and oversight, the potential of AI in task generalization and reinforcement learning, and the safety concerns surrounding their deployment.

Episode Highlights

Topics covered

Episode Highlights

Related Episodes

Ilya Sutskever (OpenAI Chief Scientist) - Building AGI, Alignment, Spies, Microsoft, & Enlightenment
Answers 383 questions
Paul Christiano - Preventing AI Takeover
Answers 383 questions
Shane Legg (DeepMind Founder) - 2028 AGI, New Architectures, Aligning Superhuman Models
Answers 383 questions
Demis Hassabis - Scaling, Superhuman AIs, AlphaZero atop LLMs, Rogue Nations Threat
Answers 383 questions
Leopold Aschenbrenner - 2027 AGI, China/US Super-Intelligence Race, & The Return of History
Answers 383 questions
Carl Shulman (Pt 1) - Intelligence Explosion, Primate Evolution, Robot Doublings, & Alignment
Answers 383 questions
Carl Shulman (Pt 2) - AI Takeover, Bio & Cyber Attacks, Detecting Deception, & Humanity's Far Future
Answers 383 questions
David Deutsch - AI, America, Fun, & Bayes
Answers 383 questions
Dario Amodei (Anthropic CEO) - $10 Billion Models, OpenAI, Scaling, & AGI in 2 years
Answers 383 questions
Holden Karnofsky - Transformative AI & Most Important Century
Answers 383 questions
Joe Carlsmith - Otherness and Control in The Age of AGI
Answers 383 questions
Eliezer Yudkowsky - Why AI Will Kill Us, Aligning LLMs, Nature of Intelligence, SciFi, & Rationality
Answers 383 questions
Steve Hsu - Intelligence, Embryo Selection, & The Future of Humanity
Answers 383 questions
Francois Chollet - LLMs won’t lead to AGI - $1,000,000 Prize to find true solution
Answers 383 questions
Sholto Douglas & Trenton Bricken - How to Build & Understand GPT-7's Mind
Answers 383 questions

Dexa/Dwarkesh Patel

John Schulman (OpenAI Cofounder) - Reasoning, RLHF, & Plan for 2027 AGI

Topics covered

Popular Clips

Diverse Labelers, Model Generalization

Human Oversight Dilemma

Model Preferences

Data Wall Debate

Coordination for AI Safety

Model Drives and Reasoning

AI Website Interaction

AI Model Alignment

Building Aligned AI

Model Intelligence Scaling

Learning Strategies Evolution

Accelerated AI Deployment

Language Model Development

Chat GPT Fine Tuning

Active Learning Models

Episode Highlights

Future AI Applications

AI Capabilities Advancement

AI Safety Concerns

Related Episodes

Ilya Sutskever (OpenAI Chief Scientist) - Building AGI, Alignment, Spies, Microsoft, & Enlightenment

Paul Christiano - Preventing AI Takeover

Shane Legg (DeepMind Founder) - 2028 AGI, New Architectures, Aligning Superhuman Models

Demis Hassabis - Scaling, Superhuman AIs, AlphaZero atop LLMs, Rogue Nations Threat

Leopold Aschenbrenner - 2027 AGI, China/US Super-Intelligence Race, & The Return of History

Carl Shulman (Pt 1) - Intelligence Explosion, Primate Evolution, Robot Doublings, & Alignment

Carl Shulman (Pt 2) - AI Takeover, Bio & Cyber Attacks, Detecting Deception, & Humanity's Far Future

David Deutsch - AI, America, Fun, & Bayes

Dario Amodei (Anthropic CEO) - $10 Billion Models, OpenAI, Scaling, & AGI in 2 years

Holden Karnofsky - Transformative AI & Most Important Century

Joe Carlsmith - Otherness and Control in The Age of AGI

Eliezer Yudkowsky - Why AI Will Kill Us, Aligning LLMs, Nature of Intelligence, SciFi, & Rationality

Steve Hsu - Intelligence, Embryo Selection, & The Future of Humanity

Francois Chollet - LLMs won’t lead to AGI - $1,000,000 Prize to find true solution

Sholto Douglas & Trenton Bricken - How to Build & Understand GPT-7's Mind

John Schulman (OpenAI Cofounder) - Reasoning, RLHF, & Plan for 2027 AGI

Topics covered

Popular Clips

Episode Highlights

Future AI ApplicationsJohn Schulman explores the future of AI assistants as proactive collaborators and the potential for AI systems to run entire companies. He discusses the importance of human oversight and regulation in maintaining trust and alignment.

Future AI Applications

AI Capabilities AdvancementJohn Schulman explores how AI models generalize across tasks and the potential of long horizon reinforcement learning. He highlights the rapid progression of AI capabilities and the importance of evaluating models for safe deployment.

AI Capabilities Advancement

AI Safety ConcernsJohn Schulman addresses the human oversight dilemma in AI processes and the challenges of coordinating AI deployment pauses. He emphasizes the need for regulatory measures and gradual, incremental releases to ensure safety and alignment.

AI Safety Concerns

Related Episodes