Training Data

Lin Qiao, founder and CEO of Fireworks and former head of the PyTorch team at Meta, shares insights on the evolving landscape of generative AI. She discusses how her platform aims to democratize access to AI with fast, cost-effective inference using smaller models. Lin explains the challenges B2C companies face with latency and operational costs. She also predicts the convergence of open and closed-source models and highlights the importance of simple API access for diverse AI applications. Her vision could transform how businesses utilize AI technology.

Thomas Dohmke, CEO of GitHub, leads the conversation on the revolutionary impact of GitHub Copilot, born from OpenAI’s GPT technology. He shares how a small team transformed Copilot from simple autocomplete to a comprehensive tool that boosts developers' productivity. The discussion highlights the need for secure software supply chains and the growing accessibility of coding. Thomas envisions a future where a billion developers can harness these AI tools by 2030, democratizing software creation and enhancing collaboration across the industry.

Joe Spisak is the head of Product Management for Generative AI at Meta, leading the team behind the recently launched Llama 3.1 405B model. He discusses how this model, trained on 15 trillion tokens, enhances zero-shot tool use and aids in generating synthetic data. Spisak believes frontier models will commoditize, benefiting startups and promoting open-source innovation. The conversation also touches on the integration of coding to improve AI reasoning, the advantages of smaller models, and the balance between open and closed systems in AI development.

Sebastian Siemiatkowski, CEO of Klarna, discusses how their AI-powered assistant revolutionized customer service, replacing 700 reps. He explores the impacts of AI on society and the arts, emphasizing an experimental AI mindset at Klarna.

AI expert Misha Laskin discusses the current state of AI agents and the need to unlock agentic capability for LLMs. He shares insights on building reliable models with Reflection AI and leveraging search capabilities of RL. Topics include leaving Russia, his journey into AI, and evolution of agents from AlphaGo to universal agents.

Microsoft CTO Kevin Scott discusses the enduring trend of scaling in AI, emphasizing the shift to inference over training as models improve. He touches on the future business models for training data and speculates on ad units for autonomous agents. Scott's short-term pessimism contrasts with his long-term optimism for industry growth and Microsoft's AI platform.

Mike Knoop, co-founder of Zapier, discusses launching the ARC Prize to stimulate AGI development. The podcast covers the importance of new benchmarks for AI, efficiency in acquiring skills, differences between foundation models and ArcGIS, techniques for improving puzzle solving, and advancing AI systems for reliability and innovation. The episode also explores innovative competitions, open-source collaboration, and the potential of AGI to revolutionize AI.

Matan Grinberg and Eno Reyes of Factory discuss creating droids for software development, leveraging AI to enhance engineering tasks. They emphasize practicality for engineers, autonomy in software, and the importance of fast execution and obsession in building a successful company.

Harrison Chase of LangChain focuses on enabling the orchestration layer for AI agents. He discusses the evolution of agents, impact on work and creativity, challenges in AI models, and solutions in LangChain. The conversation explores the role of cognitive architectures, fine tuning, and leadership reflections in the AI field.

Delve into the world of generative AI with leading AI builders and researchers as they discuss scaling laws, reasoning in LLMs, inference compute, and the future implications of AI technology