DataFramed #334 The State of Data & AI with Tom Tunguz, VC at Theory Ventures
40 snips
Dec 1, 2025 
Tomasz Tunguz, General Partner at Theory Ventures and a data-driven venture capitalist, discusses the rapid evolution of AI and its implications. He delves into how AI models like Gemini 3 are breaking performance limits and reshaping workflows through enhanced automation. The conversation covers the changing job market, urging professionals to adapt and embrace AI's capabilities. With emerging roles like forward-deployed engineers, Tunguz highlights the necessity of continuous product-market fit revalidation and the significant impact of venture capital on AI innovation.
AI Snips
Chapters
Books
Transcript
Episode notes
Pre-Training Limits Were Broken
- Gemini 3 shattered the belief that pre-training had hit scaling limits and reopened major performance gains.
- Combined model improvements and faster chips will drive real productivity gains across industries.
Post-Training And Synthetic Data Power Gains
- Post-training techniques like reinforcement learning and synthetic data are powering recent leaps in model performance.
- Synthetic data can create a virtuous cycle but risks model collapse if overused.
Tool-Calling Improvements Unlock Agents
- Tool-calling reliability has been the historical bottleneck for multi-step agents but recent models show dramatic practical improvements.
- Better tool calling enables longer multi-step workflows across CRM, ticketing and enrichment tasks.
