Does AI have human-level intelligence? (Nature Comment) (www.nature.com)

A recent commentary published in *Nature* argues that current AI models, particularly large language models (LLMs) like OpenAI's GPT-4.5, should be recognized as exhibiting artificial general intelligence (AGI). The authors highlight that these models have demonstrated the capability to perform complex cognitive tasks—such as winning gold medals in mathematical competitions, collaborating on scientific research, and engaging in human-like conversations—showing a breadth and depth of intelligence akin to that of humans. Notably, GPT-4.5 was found to pass a Turing test, being perceived as human 73% of the time, underscoring significant advancements in AI's abilities. The commentary addresses ongoing skepticism in the AI/ML community regarding the existence of AGI, suggesting that definitions commonly used in discourse are ambiguous or overly stringent. It emphasizes that current LLMs fulfill criteria for general intelligence by successfully tackling diverse cognitive challenges without needing to emulate human-like cognition, a perspective that shifts the focus from fears about societal disruption to a more nuanced understanding of machine intelligence. The authors call for the AI community and policymakers to reconsider their assessments of AI capabilities in light of these developments, highlighting the need for a clear understanding of intelligence that transcends anthropocentric biases.