Playing the long game
summary
Argues that LLM progress is messy not linear: scaling laws conflict, inference tricks matter, and DeepSeek R1 trains for about a tenth of GPT‑4 cost, which reshapes economics. Notes open‑weight models trail frontier models by ~6–12 months, and urges diversification of providers and guardrails. Cites concrete numbers: LLM‑assisted teams show 15–20% productivity gains (Stanford) and AI‑skilled wages jump from ~25% to ~56% premium (PWC 2023→2024). Recommends 70/20/10 investment focus on people, infrastructure, then models.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 3:42AI model history
- 10:57Diminishing returns in LLMs
- 16:11AI job market impact
- 19:24Software engineering today
- 22:36Organizations adapting to AI
- 34:16Code review bottleneck
- 39:56Organizations changing with AI
- 44:55Final advice
related talks
12mKeynote: How LLMs Have Changed the Way Organizations Do AI- 12mKeynote: How LLMs Have Changed the Way Organizations Do AI
24mHarnessing the Power of LLMs in Advanced Analytics
19mLLMs reasoning and agentic capabilities over time
25mWhy LLMs Still Fail: Reasoning, Robustness, and Safety
30mLLMs: Beyond the Hype - A Practical Journey to Scale
31mThe AI Revolution Will Not Be Monopolized: How open-source beats economies of scale, even for LLMs
27mProductionizing Open Source LLMs