Subvolts take Understand AI's inner workings for safer, trustworthy artificial general intelligence. Jul 10, 2026 - Developers
- Learn about techniques for understanding and auditing AI models, crucial for building robust systems.
- Business
- Understand the importance of AI safety and trustworthiness research for future AI deployment.
This episode offers a deep dive into AI interpretability, a vital research area for all major AI labs.
Understanding the inner thoughts of AI
Audio streams from Hannah Fry. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Google DeepMind: The Podcast features Neel and host Hannah Fry discussing the "neuroscience" of AI. They explore understanding complex AI intelligence without a manual, discovering structures within networks like spare autoencoders, and the limits of examining AI internals. The episode emphasizes why interpretability is crucial for building safe, aligned, and trustworthy AI as we approach AGI.
This episode explains how scientists are trying to understand how artificial intelligence "thinks" without a manual. They look for patterns inside AI systems to make sure they are safe and reliable as they get smarter.
This episode delves into the critical field of AI interpretability, explaining how researchers are trying to understand AI's inner workings. It's for anyone concerned about AI safety and trustworthiness, especially as AI capabilities advance towards AGI. This research directly supports Google DeepMind's stated goals and is a fundamental challenge for all frontier labs aiming to build responsible AI systems.
- HostHannah Fry
- GuestNeel
- Episode Title: Understanding the inner thoughts of AI
- Show: Google DeepMind: The Podcast
- Length: 53 minutes
- Guest: Neel
- Research Area: Neuroscience of artificial intelligence
- Key Concept: Spare autoencoders
Neel and his team are trying to do something phenomenally difficult: understand an intelligence that didn't come with a manual. Together, they explore the cutting-edge "neuroscience" of artificial intelligence—revealing the surprising, elegant structures being discovered inside these networks (like spare autoencoders), the inherent limits of looking under the hood, and why interpretability is absolutely essential if we are to build safe, aligned and trustworthy AI as we move towards AGI. Learn more about this area of research via https://deepmind.google/
Timecodes
- 00:00 Introduction
- 02:41 Motivation for interpretability research
- 04:01 Mechanistic interpretability
- 08:14 Chain of thought monitoring
- 18:14 Interpretability techniques
- 35:00 Auditing models for safety
- 48:53 What comes next for interpretability
Please leave us a review on Spotify or Apple Podcasts if you enjoyed this episode. We always want to hear from our audience whether that's in the form of feedback, new idea or a guest recommendation!
Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.
- spare autoencoders
- Neel
- Specific structures discovered inside AI networks.
- interpretability
- Neel
- The ability to understand how an AI system works internally.
- AGI
- Neel
- Artificial General Intelligence, AI with human-like cognitive abilities.
44 minCan AI help us better predict the weather?
WithPeter Battaglia
45 minThe mathematics of AI uncertainty
WithZoubin Ghahramani
50 minThe Future of AI Infrastructure with CoreWeave
WithCorey Sanders
41 minWhat racing reveals about working with AI - Episode 22
WithJoyce Ruffell, Chase Holden
Questions people ask
- What is the main goal of the research discussed?
- The team is trying to understand artificial intelligence that did not come with a manual, exploring the "neuroscience" of AI.
- Why is interpretability important?
- Interpretability is essential for building safe, aligned, and trustworthy AI as we move towards AGI.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Hannah Fry.