The False Promise of Safe AI - The Atlantic

The Atlantic published an article headlined "The False Promise of Safe AI" addressing the alignment problem in artificial intelligence. The article reports that three months before ChatGPT's launch, OpenAI researchers wrote that "there is currently no known indefinitely scalable solution to the alignment problem." The quoted sentence refers to a solution that would scale indefinitely, a formulation the article reproduces directly.

The article states that four years later the alignment problem remains unsolved. It also states that there is no consensus about what the problem even is. The piece sets the pre-launch assessment by OpenAI researchers against the position four years afterwards, and reports no resolution in the interval. The comparison offered is between a research note written before ChatGPT existed publicly and the state of the field after four years of subsequent work.

The headline characterises the promise of safe AI as false, and the article's stated findings — an unsolved problem, and no agreed definition of it — are presented in support of that framing. The excerpt does not name the OpenAI researchers responsible for the quoted line, identify the document in which it appeared, or describe any alignment method developed in the four years that followed. Nor does it indicate whether any of those researchers later revised that assessment. The only quantities given are the four years between the research note and the article, and the three months between the note and ChatGPT's launch.