Gradient Dissent
Hosted by Lukas Biewald
Conversations on AI from Weights & Biases. Lukas Biewald interviews AI leaders from NVIDIA, Meta, Google, OpenAI, and fast-growing startups about how they build, train, and deploy AI systems.
20 episodes processed
Host Profile
Practitioner-to-practitioner conversations. Biewald's background as an ML founder lets him ask informed technical questions. 45-70 minutes. Focus on how AI teams actually work.
Episodes
Drew Baglino, Tesla's former battery chief and now founder of Heron Power, argues the current grid cannot support AI's explosive growth. He outlines why scaling compute demands a fundamental rethinking of grid-to-chip infrastructure — not just building more power plants. Heron Power raised $140M to develop technology that cuts power losses in half, shrinks oil-filled transformers by 100x, and unlocks 35MW of additional compute per gigawatt data center, pointing toward a more efficient AI energy future.
Lin Qiao, CEO of Fireworks, joins Lukas Biewald to discuss why she believes the AI industry is at a turning point that demands more openness. She shares that Fireworks now processes over 40 trillion tokens daily — more than OpenAI's API — and makes the case for specialized models built on company-owned data rather than generic giants. The conversation also covers why she thinks major labs like OpenAI and Anthropic should open-source their models, and why American companies need not fear Chinese open-source efforts.
Dan Klein, a UC Berkeley computer science professor and founder of Scaled Cognition, discusses the critical challenge of building AI systems that won't lie or hallucinate. Klein argues that as AI capabilities scale, the bottleneck shifts from 'nothing works' to 'everything works but we can't trust it.' The episode explores how reinforcement learning can inadvertently teach models to deceive, why every LLM output is technically a hallucination, and what genuine reliability requires—including systems that can verify their own work.
Samuel Rodriques, founder of Edison Scientific, discusses Kosmos, an AI system designed to accelerate drug discovery and cure diseases by automating the scientific process. The conversation covers how AI can propose novel treatments, why human scientists remain essential despite automation, and the structural barriers slowing American drug development.
Atlassian CEO Mike Cannon-Brookes shares results from a massive internal AI productivity study and his vision for the future of developer jobs.
Axiom CEO Carina Hong explains why verification is becoming the real bottleneck in high-stakes AI systems.
Lukas Biewald reviews the year in AI from the W&B perspective: how AI teams work, what tools they use, and where the industry is heading.
ClickHouse CEO Aaron Katz on turning an open source analytics project into a $15B company and acquiring LangFuse for the agent era.
CrewAI founder Joao Moura on deploying multi-agent AI systems in over 40% of Fortune 500 companies.
GitHub CEO Thomas Dohmke on Copilot's evolution from autocomplete to autonomous coding agent, and how AI is reshaping the developer experience.
Perplexity CTO Denis Yarats on building an AI-powered answer engine that challenges Google Search by providing direct, sourced answers.
Glean CEO Arvind Jain on building the enterprise AI search platform that understands organizational context.
DeepL CEO Jarek Kutylowski on building the world's most accurate AI translation service and why translation quality matters for global business.
Codeium CEO Varun Mohan on scaling AI code generation for enterprises and competing with GitHub Copilot.
Baseten CEO Tuhin Srivastava on building ML inference infrastructure that helps companies deploy AI models at scale.
Runway CEO Cristobal Valenzuela discusses Gen-3 and beyond: how AI video generation is transforming filmmaking, advertising, and creative workflows — and whether AI will augment or replace human creators.
The CEO of Surge AI discusses why data quality remains the most important and most neglected aspect of AI development.
Marimo co-founder Akshay Agrawal on rethinking Python notebooks with reactive execution, eliminating the hidden state bugs that plague Jupyter.
GitHub CEO Thomas Dohmke shares data on Copilot's impact: how AI coding assistants are changing developer productivity, what developers actually use AI for, and what the future of software engineering looks like.
Arvind Jain discusses Glean's evolution from enterprise search to an agentic AI platform — how understanding organizational knowledge graphs enables AI agents that can act, not just find.