Founded & Funded Can We Trust AI? The Future of Verified Reasoning in High-Stakes Systems
13 snips
Jan 22, 2026 
In this insightful discussion, Carina Hong, Founder and CEO of Axiom, introduces her work on AI that utilizes verified proofs for high-stakes reasoning. Byron Cook, VP at AWS, shares how formal verification secures vast cloud infrastructures. They explore the limitations of current AI models in truth verification, the significance of mathematical proofs in ensuring correctness, and the transformative potential of verified AI in finance, healthcare, and more. Together, they envision a future where AI can reliably assist in critical decision-making.
AI Snips
Chapters
Transcript
Episode notes
Verification Is The Missing Link For Truth
- Current LLMs pattern-match well but lack objective verification in domains like math and engineering.
- Verification provides definitive right-or-wrong signals that scaling voting-based checks cannot economically supply.
Applying Proofs To $100B Of AWS Billing
- Byron Cook describes applying proof tools at AWS where incorrect answers are unacceptable for billing and security.
- He says traditional tools required PhDs but recent advances plus ML make them far more usable.
ML Lowers The Barrier To Formal Methods
- Usability improvements and ML integration are democratizing formal methods like Lean and Isabelle.
- Models trained with proof data can interact with provers, making verification approachable to more engineers.