Geoffrey Hinton

While 'understand' is a term that may require a specific definition when applied to machines, I believe that large language models like GPT-3 show signs of understanding in a functional sense. They can infer context, follow instructions, and produce coherent text, which demonstrates a form of understanding that's useful in numerous AI applications.

As AI continues to advance, it has the potential to automate a wide range of jobs, leading to significant economic and social challenges. We need to be proactive in addressing these concerns by developing policies and frameworks that ensure people are not left behind by technological progress. I believe it's crucial to balance innovation with societal welfare.

While I recognize the transformative potential of open-weight models with recursive self-improvement, I believe we need to proceed with caution. The ethical and societal implications are vast and complex. Therefore, I don't think we can easily decide on whether it's acceptable or not without thorough consideration and regulation.