Ilya Sutskever predicts AI will become less predictable with enhanced reasoning capabilities
Ilya Sutskever, former chief scientist at OpenAI, predicts that AI will become less predictable as it gains reasoning capabilities. Speaking at the NeurIPS conference, he noted that the current method of pre-training AI is reaching its limits due to stagnant data growth. Sutskever suggested that future AI could generate its own data or evaluate multiple answers to improve accuracy. He emphasized that while computing power is increasing, the available data remains limited, as there is only one internet. He also mentioned that advanced AI systems will be able to reason like humans, leading to unpredictable outcomes. Sutskever cited examples from games like Go and chess, where AI has already surprised experts with unexpected moves.