Evaluating AI Opportunities and Data

AI and product management | Marily Nika (Meta, Google)

Topic 6 of 16

Published Feb 5, 2023 · Clip 13:22–18:35 (5:14)

Marily warns against the 'shiny object trap'—don't pursue AI without a clear problem; focus on high-level solutions first. For MVPs, fake AI features in Figma to validate, not build real models early. Data needs vary: minimal for simple classification (e.g., 15-20 cat/dog photos), thousands for complex NLP. For startups, use existing APIs like GPT over building custom unless having unique data; big companies train custom for differentiation. PMs decide launch quality thresholds, e.g., 70-80% accuracy.