The LLM Revolution: From Inception to the Future of Thinking Machines
summary
Covers concrete claims, experiments and numbers: shows the scaling law that predicts performance from data and parameters, cites models at half a trillion to 1.5 trillion parameters, and demonstrates failures like depth‑map reversals, visual illusion misjudgments and the reversal curse with machine‑caption transcripts. Walks through fixes and failing cascades — model collapse demonstrated with MNIST and toy prompts — and engineering gains such as speculative sampling gives 2–3× practical speedups plus quantization and MoE tradeoffs, with demos and why they do or don’t work.
AI-written summary · about
chapters (8)
- 0:00Welcome
- 1:32LLMs love and hate
- 6:54Accidental LLM discovery
- 14:23LLMs fundamental limits
- 20:07Why LLMs hallucinate
- 29:22Failed fixes cascade
- 31:07Scale to sober rebuild
- 38:17Q&A
related talks
35mReading the Mind of an LLM
29mThe Evolution of LLMs: From Imitation to Reasoning
12mKeynote: How LLMs Have Changed the Way Organizations Do AI
01h 49mAI engineering with open access LLMs that lie, curse, and steal
42mOpen Source LLMs — Powering AI Accessibility and Innovation- 12mKeynote: How LLMs Have Changed the Way Organizations Do AI
27mThe Cutting Edge of AI: From LLMs to Multimodal Intelligence
46mPlaying the long game