Serverless AI Workflows

Eric discusses the suitability of serverless for steady inference traffic and the challenges of using it for interactive training workflows. He emphasizes the importance of observability in training jobs and the limitations of serverless for batch processing tasks.