Subvolts take Understand AI's hidden reasoning with Anthropic's J-space and explore AI's future through superforecasting. Jul 9, 2026 - Developers
- Learn about Anthropic's J-space research for understanding AI models and discuss AI hardware advancements.
- Business
- Understand how AI interpretability and forecasting research could impact future AI governance and enterprise deployment.
This episode covers significant AI research from Anthropic and discusses key areas like interpretability and forecasting.
AI:AM Highlights: Exploring the J-Space, AI Superforecasters, SambaNova's Chips, & LTX Video Gen
Audio streams from Erik Torenberg, Nathan Labenz. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
This AI:AM highlights episode features Nathan Labenz and Prakash Narayanan discussing Anthropic's "global workspace" paper, including J-space and J-lens concepts for understanding AI reasoning. They also cover AI writing detection, AI superforecasting, open world models, and AI hardware. The central theme is whether current methods can keep advanced AI systems governable as their reasoning becomes more complex.
This episode explores new ways to peek inside AI's 'brain' to see how it thinks, using concepts like J-space. It also looks at predicting AI's future performance and how to build better AI hardware, all to ensure we can trust AI as it gets smarter.
This episode offers insights into cutting-edge AI interpretability research from Anthropic, exploring how we might understand complex AI thought processes. It's for AI researchers, developers, and anyone concerned with AI safety and governance. The discussion on J-space and AI superforecasting connects directly to frontier labs' efforts to build more reliable and understandable AI systems, a key challenge for OpenAI, Google Gemini, and Anthropic.
- HostNathan Labenz
- GuestDan SchwarzFutureSearch
- GuestZeev Farbman
- GuestKunle Olukotun
- Episode Title: AI:AM Highlights: Exploring the J-Space, AI Superforecasters, SambaNova's Chips, & LTX Video Gen
- Show: The Cognitive Revolution (Anthropic)
- Length: 127 minutes
- Topic 1: Anthropic's 'global workspace' paper, J-space, and J-lens claims
- Topic 2: AI superforecasting with Dan Schwarz of FutureSearch
- Topic 3: AI hardware discussion with Kunle Olukotun
Nathan Labenz and Prakash Narayanan lead this AI:AM highlights episode with a live, hosts-only exploration of Anthropic’s “global workspace” paper, including the J-space and J-lens claims about readable concepts inside language models and the limits of what current probes can see. The episode then moves through Prakash’s AI Engineer World’s Fair field notes, Pangram AI-writing detector experiments, Dan Schwarz of FutureSearch on past-casting and AI superforecasting, Zeev Farbman on open world models, and Kunle Olukotun on the compute layer. The central stake is whether interpretability tools, forecasting benchmarks, enterprise deployment patterns, and AI hardware can make increasingly capable systems more legible and governable before their reasoning becomes too hidden to trust.
For full show notes, links, and references, read the episode page:
https://www.cognitiverevolution.ai/ai-am-highlights-exploring-the-j-space-ai-superforecasters-sambanova-s-chips-ltx-video-gen/
Sponsor:
Claude:
Claude by Anthropic is an AI collaborator that understands your workflow and helps you tackle research, writing, coding, and organization with deep context. Get started with Claude and explore Claude Pro at https://claude.ai/tcr
CHAPTERS:
(00:00) J-space paper preview
(05:37) Monitoring hidden reasoning
(13:31) Scale and critiques (Part 1)
(17:18) Sponsor: Claude
(19:09) Scale and critiques (Part 2)
(19:10) Finding hidden goals
(30:00) Anthropomorphic safety optimism
(39:02) Engineer field notes
(42:41) Detecting AI writing
(49:30) Enterprise workflow risks
(52:08) Forecasts and world models
(01:35:15) Building Q live
(01:37:24) SambaNova inference architecture
(01:42:38) AI chip taxonomy
(01:47:27) Bandwidth over capacity
(01:53:32) Testing model iterations
(01:55:35) Enforcing espoused values
(01:58:06) AI panopticon bargain
(02:01:40) Closing programming note
(02:02:29) Episode Outro
(02:05:48) Outro
PRODUCED BY:
SOCIAL LINKS:
Website: https://www.cognitiverevolution.ai
Twitter (Podcast): https://x.com/cogrev_podcast
Twitter (Nathan): https://x.com/labenz
LinkedIn: https://linkedin.com/in/nathanlabenz/
Youtube: https://youtube.com/@CognitiveRevolutionPodcast
Spotify: https://open.spotify.com/show/6yHyok3M3BjqzR0VB5MSyk
- J-space
- Nathan Labenz
- A way to visualize and understand concepts inside AI models.
- J-lens
- Nathan Labenz
- A tool or method to examine and interpret AI's internal workings.
- interpretability
- Nathan Labenz
- The ability to understand how an AI system makes its decisions.
- superforecasting
- Dan Schwarz
- Making highly accurate predictions about future events.
1 h 27AI:AM: Where's the Steering Wheel? My Agent Set the Password, Hidden Valence, Proofs Find Linux Bugs
WithCameron Berg, Everett Hildenbrandt, Grigore Rosu, Phil Maher
2 h 192 h 021 h 28AI:AM: A Level We Shouldn't Pass? Notes from The Curve + Tokens vs. Salaries & Is SaaS Cooked?
WithThomas Sohmers, swyx, Evan Miyazono, Edward Hu
Same guests, other shows
58 minDataflow Computing for AI Inference with Kunle Olukotun - #751
WithKunle Olukotun, Kunle Olukotun
49 minAnthropic's Katelyn Lesse & Angela Jiang: Building an Ecosystem, not a Walled Garden
WithKatelyn Lesse, Angela Jiang
Questions people ask
- What is J-space and J-lens?
- J-space and J-lens are concepts from Anthropic's research aiming to reveal readable concepts within language models, exploring the limits of current interpretability tools.
- What is AI superforecasting?
- AI superforecasting involves using AI and expert insights to predict future events with greater accuracy, as discussed by Dan Schwarz of FutureSearch.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Erik Torenberg, Nathan Labenz.