Eliezer Yudkowsky

If Earth experiences a sufficient rate of nonhuman manufacturing -- eg, self-replicating factories generating power eg via fusion -- to saturate Earth's capacity to radiate waste heat, humanity fries. It doesn't matter if the factories were run by one superintelligence or 20. (source)

The most likely result of building a superhumanly smart AI... is that literally everyone on Earth will die. (source)

Allowing the development of digital humans with full cognitive abilities is incredibly dangerous. The unpredictable nature of advanced AI poses existential risks that we are not prepared to handle. We must prioritize strict regulation and safety research before venturing into such uncharted territory.

Neural interfaces are not a silver bullet for AI alignment. While they might offer new ways to interact with AI, the core challenge of aligning AI’s goals with human values remains deeply complex and unsolved.

I don't believe expanding LLMs alone is sufficient for achieving AGI. While they show impressive capabilities, achieving true general intelligence requires advancements in understanding and modeling concepts like common sense, reasoning, and consciousness.