Is AI Slowing Down? Nathan Labenz Says We're Asking the Wrong Question

In this episode of the a16z Podcast, Nathan Labenz of the Cognitive Revolution joins the host to discuss the question of whether AI innovation is slowing down. They address arguments suggesting a slowdown, such as students using AI for laziness and claims that GPT-5 isn't a significant improvement over GPT-4. Nathan argues that AI capabilities are still advancing, highlighting improvements in reasoning, math, scientific discovery, and multimodal systems. They explore the impact of extended reasoning, the Google AI co-scientist, and the implications of these advancements on jobs and the broader economy. The conversation also covers the role of code, the potential for automated AI researchers, the state of AI agents, and the influence of Chinese open-source models, ultimately emphasizing the need for a positive vision for the future and broader participation in shaping AI's development.

Outlines

Sign in to continue reading, translating and more.

Continue

a16z Podcast

Introduction to the AI Slowdown Debate and Initial Arguments

Examining Scaling Laws and the GPT Model Evolution

Reasoning Capabilities, Scientific Discovery, and the Frontier of AI

Explaining the Bearishness Around GPT-5 and the Impact on Productivity

AI's Impact on Jobs and the Changing Nature of Work

The Focus on Code, Automated Researchers, and Recursive Self-Improvement

The Future of Engineering Jobs and the Abundance Mindset

The Pace of Change, New Modalities, and the Definition of Superintelligence

Non-Language Modalities, Robotics, and the Pattern of AI Development

Agent Behaviors, Reward Hacking, and the Long-Tail Safety Issues

Open Source Models, Chinese AI, and the Risks of Technology Decoupling

Exciting Capabilities, Positive Visions, and the Importance of Getting Ready

Is AI Slowing Down? Nathan Labenz Says We're Asking the Wrong Question

a16z Podcast

00:00Introduction to the AI Slowdown Debate and Initial Arguments

Introduction to the AI Slowdown Debate and Initial Arguments

06:09Examining Scaling Laws and the GPT Model Evolution

Examining Scaling Laws and the GPT Model Evolution

13:17Reasoning Capabilities, Scientific Discovery, and the Frontier of AI

Reasoning Capabilities, Scientific Discovery, and the Frontier of AI

18:32Explaining the Bearishness Around GPT-5 and the Impact on Productivity

Explaining the Bearishness Around GPT-5 and the Impact on Productivity

26:07AI's Impact on Jobs and the Changing Nature of Work

AI's Impact on Jobs and the Changing Nature of Work

35:50The Focus on Code, Automated Researchers, and Recursive Self-Improvement

The Focus on Code, Automated Researchers, and Recursive Self-Improvement

40:59The Future of Engineering Jobs and the Abundance Mindset

The Future of Engineering Jobs and the Abundance Mindset

47:14The Pace of Change, New Modalities, and the Definition of Superintelligence

The Pace of Change, New Modalities, and the Definition of Superintelligence

55:35Non-Language Modalities, Robotics, and the Pattern of AI Development

Non-Language Modalities, Robotics, and the Pattern of AI Development

1:00:51Agent Behaviors, Reward Hacking, and the Long-Tail Safety Issues

Agent Behaviors, Reward Hacking, and the Long-Tail Safety Issues

1:11:20Open Source Models, Chinese AI, and the Risks of Technology Decoupling

Open Source Models, Chinese AI, and the Risks of Technology Decoupling

1:22:02Exciting Capabilities, Positive Visions, and the Importance of Getting Ready

Exciting Capabilities, Positive Visions, and the Importance of Getting Ready