
The MAD Podcast with Matt Turck Poolside: AI for Software Development with CTO Eiso Kant
Dec 20, 2023
Eiso Kant, CTO and co-founder of Poolside, shares his journey from a young programmer to co-founding a trailblazing AI startup. He discusses Poolside's foundational model designed for software developers and emphasizes the critical role of data quality. Eiso unveils their approach to creating AI tools that leverage executable code for scalable improvement, and highlights the importance of building a vertically integrated model tailored for developers. He also reflects on the benefits of establishing the company in Europe, focusing on global talent and competition.
AI Snips
Chapters
Transcript
Episode notes
Code As A Path To Broader Intelligence
- Poolside focuses on software development as a proxy task to accelerate general AI capabilities.
- Executable source code enables programmatic evaluation and large-scale improvement loops for models.
Sequence: Assist First, Automate Later
- Start by shipping an AI pair-programming assistant before pursuing full autonomy.
- Expose your models via APIs so others can build complementary autonomous experiences.
Programmatic Rewards Scale Learning
- Reinforcement learning from code execution feedback (RLCF) uses tests and compilers as programmatic oracles.
- This enables scaling reward signals far beyond human feedback and can teach new capabilities.

