Systems like GPT-4 and beyond have the potential to reach new levels of capability, as we've seen with previous scaling. Adding orders of magnitude could unlock further capabilities, including long-term planning and high-quality AI research. The phenomenon of grokking shows that AI can rapidly improve once it understands underlying principles. The implications of this for AGI, such as Grok, chosen by X or Elon Musk, are intriguing.
UnHerd's Flo Read meets Nick Bostrom.
Hosted on Acast. See acast.com/privacy for more information.