Let's Talk GenAI
summary
Covers generative AI history from ELIZA in the 1960s to the 2017 Transformer (Attention Is All You Need) architecture and practical limits: models trained on massive corpora (Wikipedia, Twitter, web) with GPT‑4: 1.76 trillion params and reported training costs ~USD100M. Demonstrates fine‑tuning (ChatDoctor) that improves medical specificity, shows failures like Microsoft’s Tay shut down after 16 hours, highlights bias from averaged image generations across 300 faces per occupation, and reproduces hallucinations where models confidently state false facts.
AI-written summary · about
chapters (10)
- 0:00Introduction to generative AI
- 1:06GenAI historical roots
- 2:32Artificial vs generative AI
- 3:39How language models work
- 5:45Neural network black box
- 8:10Transformer revolution
- 9:35Pre-training and fine-tuning
- 12:41Parameters and training data
- 15:16Ethical risks and biases
- 17:00Biased data consequences
related talks
29mThe Rise of AI: A Reality Check on Generative AI
26mGenerative AI: Get a Grip … and a Look Under the Hood
15mOne Year of Generative AI: What Has Happened and What's Next?
27mChallenges of Generative AI
49mGenerative AI: Concepts, Tools, and Applications for Data Professionals
17mUnlock the power of Generative AI: From understanding its principles to utilizing its power
41mThe AI boom: what and why?
59mKeynote: The past, present, and future of AI for application developers