The Neuron: AI Explained

Megan Li, Senior Product Manager at Google Labs who builds Opal, shows a no-code visual workflow builder. Tomas Ilgic, Senior Director who leads Flow, demos AI filmmaking and scene editing tools. Jaclyn Konzelman, Director behind Mixboard, walks through AI-powered visual ideation and mood boards. They demo creating visuals, video clips, and chained AI apps, and discuss integration, editing workflows, and future directions.

Eno Reyes, Co-founder and CTO of Factory AI and leader in autonomous coding agents, explains Droids: fully autonomous agents that edit real codebases, run tests, and plug into dev workflows. He discusses context compression research, what makes a codebase agent-ready, model routing strategies, long-session handling, and safety guardrails for real-world autonomy.

Bowen Baker, OpenAI research scientist focused on interpretability and safety, joins to discuss how models plan, hide, and sometimes cheat. He describes reward-hacking examples and why watching chain-of-thought traces can catch problems earlier. The conversation covers monitorability limits, the tradeoff between transparency and performance, and risks of models learning to obfuscate their thinking.

Darin Patterson, VP of Market Strategy at Make, dives into the intricacies of AI agents and automation. He explains the importance of a solid automation foundation before deploying AI agents. Discover how to choose between deterministic workflows and AI agents, and learn about the role of Model Context Protocol in modern automations. Darin also highlights the significance of orchestrating these agents as a new management skill, and shares real-world examples of AI applications in marketing and finance.

IBM's new Granite 4.0 model represents a shift towards AI as a reliable tool rather than a friendly entity. The discussion highlights hybrid architectures that enhance efficiency while ensuring transparency and safety in enterprise settings. With activated LoRA adapters, models can switch functions dynamically, making AI adaptable for various tasks. The focus on treating generative models as computational engines is intriguing, alongside the emphasis on making AI boring and dependable for real-world use, aiming for safe, practical applications.

Zuzanna Stamirowska, CEO and cofounder of Pathway, dives into the revolutionary BDH architecture designed for continual learning and brain-like reasoning. She explains why current AI lacks memory and how BDH introduces real-time context and adaptability. Discover how this architecture fosters genuine learning, allowing models to 'get bored' and strengthen connections. Zuzanna emphasizes that reasoning is the core of intelligence, paving the way for significant innovations in fields like Formula 1 and beyond.

Karina Hong, a Stanford dropout and founder of Axiom Math, is on a mission to create a self-improving mathematical reasoning AI. She reveals how her team tackled a 130-year-old problem and disproved a longstanding conjecture, showcasing the power of math as a foundation for fields like coding and finance. Karina discusses the challenges of auto-formalizing proofs and highlights Axiom's vision of an IDE co-pilot that aids researchers and traders alike. The exciting intersection of neuroscience and AI is explored, promising a revolutionary leap in mathematical discovery.

Nick Talken, CEO of Albert Invent, shares his journey from a trailer lab to revolutionizing R&D with AI. The conversation reveals how their AI model, trained on 15 million molecular structures, accelerates research at major companies like Kenvue, reducing timelines from three months to just two days. Nick discusses the unique challenges of chemistry for AI applications, the importance of domain-specific models, and the role of scientists in guiding AI. He also highlights the significance of data sharing and the potential for democratizing materials invention.

Sam Liang, Co-founder and CEO of Otter.ai, previously developed the 'blue dot' for Google Maps. In this conversation, he shares how Otter evolved from a simple transcription tool to an AI-driven meeting agent. Sam discusses the necessity of capturing meeting data to improve ROI and prevent knowledge loss. He introduces the Otter Workspace, which organizes meetings like Slack channels and explains how AI can proactively contribute during calls. The importance of privacy controls and the technical challenges of voice understanding are also key highlights of their chat.

Pavan Davuluri, Corporate VP of Microsoft's Windows + Devices, shares his insights on transforming Windows into an AI-native platform. He discusses the deep integration of Copilot into Windows and the balancing act between powerful automation and user control. Pavan presents a future where cloud and on-device AI coalesce, enhancing familiar workflows with innovations like voice interaction and intelligent troubleshooting. He emphasizes designing with user value in mind, all while reflecting on Windows' rich 40-year legacy and its evolution towards a hybrid AI computing experience.