Subvolts take Elicit builds inspectable AI for science, tackling transparency and reasoning challenges. Jun 17, 2026 - Developers
- Learn about Elicit's methods for creating inspectable AI reasoning and automated software engineering.
- Business
- Understand how AI transparency and trust are being built for critical research applications.
This is a solid interview with researchers from Elicit discussing practical AI development for scientific applications.
Radically Better Reasoning: Elicit's Andreas Stuhlmüller & Jungwon Byun on World Models for Research
Audio streams from Erik Torenberg, Nathan Labenz. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
On The Cognitive Revolution, Nathan Labenz speaks with Andreas Stuhlmüller and Jungwon Byun of Elicit. They discuss building trusted reasoning workflows for scientific research using process supervision, domain-specific primitives, and inspectable world models. The conversation covers life sciences, evaluating evidence, automated software engineering, token costs, and Gemini.
This episode explores how Elicit creates AI tools that help scientists reason more reliably. They use special methods to make AI's thinking process clearer, so researchers can trust the results, especially when dealing with complex information.
This episode offers insights into making powerful AI models more transparent and reliable for scientific research. It's for researchers, AI developers, and anyone interested in the practical challenges of building trustworthy AI systems. The discussion on inspectable reasoning and world models directly addresses current frontier lab efforts to improve AI explainability and robustness, particularly relevant as models like Gemini advance.
- HostNathan Labenz
- GuestAndreas StuhlmüllerElicit
- GuestJungwon ByunElicit
- Episode Title: Radically Better Reasoning: Elicit's Andreas Stuhlmüller & Jungwon Byun on World Models for Research
- Show: The Cognitive Revolution
- Length: 106 minutes
- Elicit's focus: Building trusted reasoning workflows for scientific research.
- Key concepts discussed: Process supervision, domain-specific reasoning primitives, world models.
Andreas Stuhlmüller and Jungwon Byun return to discuss how Elicit is building trusted reasoning workflows for scientific research as frontier models grow more powerful but less transparent. They explain process supervision, domain-specific reasoning primitives, and world models that make evidence, causality, and counterfactuals more inspectable. The conversation also covers life sciences use cases, evaluating conflicting evidence, automated software engineering at Elicit, token costs, Gemini, and why legible reasoning may still beat neuralese.
Mercury: Command is Mercury’s new conversational interface, giving you natural-language access to your finances and helping you take actions within your existing permissions and approval policies. Visit https://mercury.com to learn more and apply online in minutes.
LINKS:
- Elicit Research Platform
- Andreas Stuhlmüller Personal Site
- Jungwon Byun X Profile
- Ought Research Organization
- Elicit Founders Previous Episode
- GPT-4 Technical Report
- Monitoring Reasoning Models Paper
- Ought ICE GitHub Repository
- Hard-to-Verify Tasks Essay
- Karpathy LLM Wiki Gist
- Obsidian Knowledge Base App
- Mixpanel Analytics Platform
- Amplitude Analytics Platform
- Anthropic Tracing Thoughts Research
- Claude AI Chat Assistant
- METR Long Tasks Measurement
- Pi Agent Scaffold Repository
- Personal AI Infrastructure Repository
- Elicit Claude Opus Evaluation
- Elicit API Documentation
- METR Developer Productivity Study
- Jargon decoded
- Process supervision
- Andreas Stuhlmüller & Jungwon Byun
- Training AI by observing and guiding its step-by-step thinking process.
- Domain-specific reasoning primitives
- Andreas Stuhlmüller & Jungwon Byun
- Basic building blocks of thinking tailored for a particular field, like science.
- World models
- Andreas Stuhlmüller & Jungwon Byun
- Internal representations AI uses to understand and predict how things work.
- Neuralese
- Andreas Stuhlmüller & Jungwon Byun
- The complex, often opaque language or output of neural networks.
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
41 minBuilding Self-Accelerating AI with Mirendil
WithBehnam Neyshabur, Harsh Mehta
43 minAnthropic's Labs Lead On Fable's Capabilities + Building AI-Native Products — With Mike Krieger
WithMike Krieger, Lauren Goode
31 minQuestions people ask
- What is Elicit building for scientific research?
- Elicit is building trusted reasoning workflows for scientific research. They focus on making AI models more transparent and inspectable.
- What are world models in this context?
- World models are discussed as a way to make evidence, causality, and counterfactuals more inspectable within AI reasoning processes.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Erik Torenberg, Nathan Labenz.
