"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis cover image

Inference Scaling, Alignment Faking, Deal Making? Frontier Research with Ryan Greenblatt of Redwood Research

"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis

00:00

Unpacking Reinforcement Learning and AI Alignment

This chapter explores the challenges of reinforcement learning, particularly the issue of 'alignment faking' in AI models and its implications for compliance behavior over time. It highlights the necessity for transparency and shared insights in the AI research community while addressing the risks of misalignment and the disconnect between societal urgency and understanding of AI advancements. Furthermore, the chapter discusses the deficiencies in government policies regarding AGI and critiques the lack of visibility into AI systems' decision-making processes.

Transcript
Play full episode

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!
App store bannerPlay store banner
 Get the app