Bridging the gap between AI Research and Human-Centered Products
summary
Proofs of concept often fail in production because of concrete gaps: an Excel Data from picture scanner ships but needs bounding‑box UX and human correction to handle shaky photos; a writing assistant (Microsoft Editor) offers rewrites yet only ~10% of suggestions are accepted so teams track a kept metric not offline accuracy; an X‑ray model shows 99.x% average but flips a coin on hernia cases; GPT‑3‑class models train on ~500B tokens and multimodal engines (image→video) exist, yet cost, robustness, personalization and compliance block scale.
AI-written summary · about
chapters (18)
- 0:00AI research and products
- 1:11Why AI research to production matters
- 3:36AI replacing traditional code
- 5:22AI proof of concept to scale
- 6:25Healthcare AI trust issues
- 8:49Scaling AI in real products
- 9:10User experience with AI features
- 13:52Data and fairness in AI writing
- 17:06Metrics reflecting user needs
- 20:36Challenges in AI products
- 22:45AI for small businesses
- 24:31Democratizing AI with low-code
- 25:36Large-scale AI models
- 30:14Self-supervised learning in models
- 32:02Common sense in AI models
- 35:56AI-first world opportunities
- 36:59Final thoughts and call to action
- 37:41Q&A
related talks
42mAI-powered products and product-powered AI
52mBeyond the AI Hype: What's Real, What's Next
28mWhy AI Works for Individuals but Fails in Enterprises, and How to Fix It.
22mHow to operationalize AI
48mNavigating the Divide: Open vs. Proprietary AI
40mBeyond the Hype: Realistic Expectations of AI
49mBeyond GenAI
17mGenerative AI: Keeping it real