20VC: Yann LeCun on Why Artificial Intelligence Will Not Dominate Humanity, Why No Economists Believe All Jobs Will Be Replaced by AI, Why the Size of Models Matters Less and Less & Why Open Models Beat Closed Models
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
Limitations of AI: Why Language-Only Training Hinders True Intelligence
AI systems may seem intelligent due to their fluency in language, but their ability to think, understand the world, and plan is limited. This is because they are trained strictly on language, which only represents a small proportion of human knowledge. Most human knowledge is non-linguistic, and the same goes for animal knowledge. The capabilities and abilities that humans take for granted, such as planning, motion, and simple tasks, cannot be replicated by current AI models. AI systems of the future that truly understand the world will not be autoregressive large language models. They will require real-world experience or simulations to develop an intuition of how the world works. Current language-only models lack a mental model of the world and cannot plan complex action sequences or use tools effectively. AI research is skeptical of prophecies that overstate the capabilities of language-only models.
Remember Everything You Learn from Podcasts
