How AI Is Built

#052 Don't Build Models, Build Systems That Build Models

Jul 1, 2025
Dive into the fascinating world of AI infrastructure with insights on building adaptive systems rather than just fine-tuned models. Explore the shift to serverless platforms and the critical role of task decomposition in model performance. Discover why inference is where the real money lies and get an understanding of GPU versus CPU processing challenges. Learn about optimizing MLOps with advanced integration patterns and improving data processing pipelines for efficiency. The conversation wraps with thoughts on enhancing community engagement in the AI development landscape.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

Serverless Enables Ambitious Scale

  • Modal enables spinning up 100 GPUs quickly, encouraging ambitious scaling in AI workloads.
  • Serverless infrastructure changes economics, letting teams rent compute for brief tasks without overhead.
ADVICE

Monetize Model Inference

  • Focus monetization on model inference integrated within software, not on training models.
  • Deliver value by combining model outputs with applications and tools like APIs and agent orchestration.
INSIGHT

Differentiation Via Distribution and Trust

  • Distribution, bundling, better UI, and access to unique data will differentiate AI offerings.
  • Trust issues limit immediate dominance by big tech despite their integrated stacks.
Get the Snipd Podcast app to discover more snips from this episode
Get the app