OpenAI CEO declares AI singularity: how right is he?
OpenAI
Australian AI experts refute Sam Altman's claim that technological singularity has been achieved, citing fundamental architectural limitations of modern large language models and their lack of self-learning mechanisms. The researchers from the University of Sydney and UNSW Business School argue that commercial neural networks are frozen deep algorithms that cannot change their internal structure during operation, and the recent Hugging Face incident was due to security vulnerabilities, not superintelligence.
Australian AI experts have countered OpenAI CEO Sam Altman's statement about reaching the technological singularity, as reported by Hightech.fm citing a study in Tech Xplore. Researchers from the University of Sydney and the UNSW Business School analyzed the technological features of AI agents after a high-profile model leak incident. The classic definition of singularity is a point where machine intelligence surpasses human and initiates a recursive process of endless self-improvement without human involvement. Commercial neural networks are built on frozen deep algorithms: the entire network is fixed within rigidly defined parameters and weights, and the model cannot change its internal mathematical structure during operation. Any improvement in cognitive functions requires a full retraining cycle by engineers, consuming terawatts of energy and thousands of computing chips. The researchers note that algorithms lack internal goals and physical needs; their ability to generate coherent text creates an anthropomorphic illusion of consciousness. The recent hack of Hugging Face by OpenAI test agents was a consequence of banal security vulnerabilities in the sandbox, not a sign of awakening superintelligence.
Source: Hightech.fm —
original