In this insightful episode, we sit down with Jeff Smith—AI/ML pioneer, PyTorch leader at Meta, and serial innovator—to unpack his journey shaping the tools and technologies powering modern AI. From spearheading PyTorch’s evolution into a community-driven backbone for research to incubating breakthroughs like ESMFold and CICERO at Meta’s FAIR Accel, Jeff reveals how open-source ethos and corporate ambition can coexist. We dive into his entrepreneurial ventures in neuromotor interfaces, conversational AI, and cloud infrastructure, exploring the messy realities of scaling AI while tackling ethical dilemmas: Should hyper-realistic voice agents disclose their non-human nature? Can democratizing models like Llama inadvertently empower bad actors? Jeff also shares lessons from his foundational books on ML systems, updates his vision for multimodal AI post-LLM boom, and reflects on guarding against hype in a field moving at light speed. Whether discussing AR’s dependency risks or the “cold start” problem in scheduling algorithms, this conversation is a masterclass in balancing technical ambition with humanity’s best interests. Tune in for a rare blend of hard-won pragmatism and bold optimism about AI’s next frontiers.