AI Snips
Chapters
Transcript
Episode notes
Google Built Transformers For Product Problems
- Google did not merely invent transformers in research; Sundar Pichai says it deployed them immediately in Search through BERT and MUM and had a proto-ChatGPT in LaMDA.
- He argues Google hesitated because its quality bar and lack of RLHF made internal systems too toxic to ship broadly then.
Latency Budgets Shape Google Product Quality
- Sundar Pichai treats latency as a core product virtue because fast responses usually signal strong technical foundations, not just polish.
- Search teams get millisecond budgets; if a change saves 3ms, only half can be reinvested and half must reach users.
Search Becomes An Agent Manager
- Sundar Pichai sees Search evolving into an agent manager rather than disappearing, with users running long tasks and information-seeking workflows through it.
- He frames AI as expansionary, not zero-sum, arguing Search and Gemini will overlap in some areas while diverging in others.