The Risks of Losing Control: A Former OpenAI Researcher Reveals AI's Secret Public Information
A former OpenAI researcher, Daniel Kokotajlo, warns that the AI industry is falling into blind competition, with companies racing to build systems they do not fully understand or control. The core challenge is the 'alignment' problem: even if AI surpasses human capabilities, researchers still cannot understand the internal decision-making mechanisms of advanced models, and the threat of superintelligence is becoming increasingly evident.