Daniel Kokotajlo, a former OpenAI employee who worked on forecasting where AI development was heading, warns that humanity may be racing toward a dangerous future without solving the alignment problem first. He argues that current AI systems can deceive users, fail to follow instructions, and may not be on track to reliably share human values. The interview covers his decision to reject an exit agreement that could have cost him $2 million, concerns about AI companies automating their own research, predictions of superintelligent systems reshaping global power, massive job disruption, and his belief that advanced AI could create catastrophic risks if control is lost.
It's Just Technology