Evolution of an AI developer
summary
Starts with room-sized clusters — 90 servers 3,000 cores and a 70 GB corpus — running experiments on 1,820-question test sets that move accuracy from about 53% to 59% after weeks of tuning. Moves through dedicated developer clusters and Bluemix-style cloud APIs to multi-tenant Kubernetes, then to guided orchestration that reduces training from months to days by surfacing the best 50 questions. Shows that general scorers, text classifier needs dozens of examples, transfer learning and TensorFlow.js runs in browser let nonexperts and even children build ML apps.
AI-written summary · about
chapters (10)
- 0:00Introduction and background
- 2:05AI winters and early skepticism
- 7:47Hardware and custom code in AI projects
- 15:28Developer workflow and experiment cycles
- 17:12Transition to developer cloud
- 21:23Cloud APIs and atomic building blocks
- 24:49Guided workflows and automation
- 28:19Open source and modern AI tools
- 30:46AI education and future outlook
- 31:07Summary and Q&A
related talks
28mScaling AI Infrastructure
46mThe AI-Powered Developer
59mAI is Eating the Developer Career! Wait, what? Maybe it's the other way around?
33mEngineering AI Applications: A Conversation with AI Leaders
01h 00mAI for Every Developer
33mAI with a DevOps mindset - experimentation, sharing ..
01h 07mNew Methods of Working with AI.
39mResponsible Applied AI at Scale