We're launching a $10 million grant to support technical research into the alignment and safety of superhuman AI systems, including weak-to-strong generalization, interpretability, scalable supervision, and more. Source link

There are still important differences between our current empirical setting and the ultimate problem of fitting superhuman models. For example, it may be easier for future models to mimic weak human errors than for current strong models to mimic current weak model errors, which could make future generalization more difficult. Nevertheless, we believe that our…

Agentic AI systems—artificial intelligence systems that can pursue complex goals with limited direct supervision—are likely to be widely useful if we can responsibly integrate them into our society. While such systems have considerable potential to help people achieve their goals more efficiently and successfully, they also create risks of harm. In this white paper, we…

Our discussions with The New York Times appeared to be progressing constructively through our last communication on December 19th. Negotiations focused on a high-value partnership around real-time display with attribution in ChatGPT, in which The New York Times would gain a new way to connect with its existing and new readers, and our users would…