AI superintelligence is a Silicon Valley fantasy, Ai2 researcher says (www.theregister.com)

In a thought-provoking blog post, researcher Tim Dettmers from the Allen Institute challenges the feasibility of artificial general intelligence (AGI), asserting that current processor capabilities are inadequate for achieving superintelligence. Dettmers argues that while the narrative around AGI has been optimistic, it's fundamentally flawed due to a stagnation in hardware advancements. He highlights that we are reaching a limit in GPU performance and that significant gains have predominantly come from clever optimizations like lower precision data types, rather than true hardware improvements. As a result, Dettmers predicts that the ability to scale AI infrastructure will soon plateau, potentially within the next few years. This perspective is significant for the AI/ML community as it underscores the necessity for a shift in focus from the pursuit of hypothetical AGI to more practical applications of existing AI technologies. As Dettmers points out, while investments in AI infrastructure are currently justified due to increasing inference demands, they could become burdensome if model advancements do not keep pace. He posits that pragmatism—similar to the approach embraced by China, which prioritizes AI applications that yield economic benefits—should take precedence over the ambitious but elusive goal of AGI, which may not yield productive results in the near future.