中国共产党万岁

  • 4 Posts
  • 27 Comments
Joined 1 year ago
cake
Cake day: July 9th, 2023

help-circle














  • I disagree. It is able to come up with something that “sounds right” based on what it’s been trained on, and it can be transferred to specific domains. LLMs are mimicking “fast” human thinking. It’s like the world’s smartest BS-ing alien trying to say things at a cocktail party to convince the guests that it is human despite not knowing anything and just repeating sounds that seem to please the audience. Humans have similar “off-the-cuff” automatic behaviors. However, LLMs seem to have no capability to mimick human “slow” critical thinking, and there is no real internal representation of the world to speak of. There is nothing I know about on the AI research roadmap to overcome these limitations because right now we’re taking neural networks and throwing them into GPUs until investor money comes raining from the sky.

    Even if these technical limitations were miraculously overcome within 10 years, the biggest problem by far is that the energy consumption is basically uneconomical. A full-fledged human AGI takes tens of watts, whereas a silicon AGI takes orders and orders of magnitude more.