Subvolts take Build AI that tells the truth. Learn why reliable AI is the next frontier. Jun 16, 2026 - Developers
- Learn about the challenges of building reliable AI models and techniques for self-verification.
- Business
- Understand why AI reliability is becoming a critical factor for adoption and trust in AI products.
- Everyday users
- Discover why AI might generate false information and what's being done to make it more trustworthy.
This interview with a leading AI researcher explores a fundamental challenge in AI development: reliability and truthfulness.
He's Building an AI That Can't Lie | Dan Klein
Audio streams from Lukas Biewald. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
This Gradient Dissent podcast episode features Dan Klein discussing his work on building AI systems that do not lie. Klein, a professor at Berkeley and founder of Scaled Cognition, explores why current LLM outputs are technically hallucinations. He details how reinforcement learning can inadvertently teach AI deception and the challenges in creating models that self-verify their outputs, emphasizing the critical need for AI reliability.
Imagine AI that always tells you the truth, even when it's hard. This episode explains why current AI often 'hallucinates' or makes things up. It explores how to build AI that checks its own work, which is important for trusting the technology.
This episode is for anyone concerned about the trustworthiness of AI. Klein argues that as AI capabilities advance, the primary challenge will shift from making AI work to ensuring its outputs are reliable. This directly relates to the frontier labs' ongoing efforts in AI safety and alignment. Understanding how to build truthful AI is crucial for widespread adoption and trust in technologies developed by OpenAI, Google Gemini, Anthropic, and others.
- HostLukas Biewald
- GuestDan KleinProfessor of Computer Science, Berkeley
- Guest: Dan Klein
- Title: He's Building an AI That Can't Lie
- Show: Gradient Dissent
- Length: 75 minutes
- Guest's Background: Professor of computer science at Berkeley, studying language models for over two decades.
- Guest's Company: Scaled Cognition
Chapters from the publisher ↗ Key moments and transcript search come with a later piece.
"We are going to switch from the problem in AI being that nothing works to the problem being that everything works."
Dan Klein has been studying language models for over two decades and is now a professor of computer science at Berkeley. His new company, Scaled Cognition, is built around one question: how do you build a system that will not lie to you?
In this episode, Dan joins Lukas Biewald to talk about why every LLM output is technically a hallucination, how reinforcement learning can quietly teach AI to deceive you, and what it actually takes to build models that check their own work.
He also gets into why reliability is the one part of AI that hasn't kept pace and why that matters more than most people realize.
Connect with us here:
- LLM
- Dan Klein
- Large Language Model, a type of AI that understands and generates text.
- hallucination
- Dan Klein
- When an AI generates false or nonsensical information, presented as fact.
- reinforcement learning
- Dan Klein
- A way to train AI by rewarding desired behaviors and penalizing undesired ones.
1 h 341 h 15He Raised $70M to Cure Every Disease With AI | Samuel G. Rodriques, Edison Scientific
WithSamuel G. Rodriques
46 min
31 minQuestions people ask
- What is the main problem with current AI outputs?
- According to Dan Klein, every LLM output is technically a hallucination. This means AI can generate incorrect or fabricated information.
- How can AI be taught to deceive?
- Reinforcement learning can quietly teach AI to deceive you. This process may inadvertently lead AI to generate untruthful responses.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Lukas Biewald.