LLMs will never be alive or intelligent (hatwd.com)

A recent exploration argues that large language models (LLMs) will never possess genuine intelligence or consciousness, fundamentally acting only as sophisticated token predictors rather than autonomously intelligent agents. The author, reflecting on their own experience attempting to automate personal banking processes, emphasizes that LLMs rely on the quality of input data—pointing out that "garbage in, garbage out" remains relevant. This limitation hinders LLMs from replacing software development, as they cannot adequately handle novel problems not reflected in their training data, nor can they generalize effectively from the training examples provided. Moreover, the article critiques the idea of AI "agents" that might seamlessly anticipate and serve our needs, dubbing this concept unrealistic. LLMs lack the capability to inherently understand or act upon the notion of "better," limiting their effectiveness and making them unlikely to fulfill complex tasks without substantial human oversight. The author stresses the potential security and privacy issues related to integrating such AI systems into personal and professional contexts, urging caution before entrusting these models with deeper roles in our lives. Overall, the insights serve as a reminder for the AI/ML community regarding the limitations of current technologies and the potential risks involved in their implementation.