The Rise of AI: A Reality Check on Generative AI
summary
Overview of concrete claims, tools and numbers: large models use embeddings and RLHF to auto-regress next-token text; GPT‑3.5 is 175B parameters and GPT‑4 is claimed at ~1T parameters; training can cost millions and, per MIT example cited, single pre‑training emits carbon comparable to several car lifespans; McKinsey estimates 15–40% new use cases and 35–70% productivity gains in some functions. Demos include prompt‑injection and adversarial examples that unlock harmful outputs; detection of deepfakes is still ~50%.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 2:30Talk purpose and scope
- 7:55AI vs ML fundamentals
- 13:16Generative AI and GPT models
- 23:18Economic impact of generative AI
- 27:14Future of AI and big tech
- 30:47Generative AI risks
- 43:41Trustworthy AI systems
- 45:44Conclusion
related talks
52mBeyond the AI Hype: What's Real, What's Next
26mGenerative AI: Get a Grip … and a Look Under the Hood
45mHow Generative AI Works: LLMs and RAG Under the Hood
17mUnlock the power of Generative AI: From understanding its principles to utilizing its power
15mOne Year of Generative AI: What Has Happened and What's Next?
29mHow AI Models Get Smarter
50mKeynote: After the AI Hype – What’s Real, and What’s Next
01h 00mAI A-Z