Keller Jordan argues that pausing AI progress could increase long-term risk

Summary

Keller Jordan presents a hypothetical argument about how policy choices could affect AI risk. He says that immediately pausing algorithmic progress at AI labs while allowing hardware companies to keep expanding compute could reduce the number of visible AI accidents over the next five years. In his view, fewer incidents would make the public and governments less alarmed, while a large stockpile of compute continued to accumulate. If algorithmic progress were later restarted, he argues, the combination could enable a sudden and unexpected capability jump, or “FOOM,” that would be difficult to control. Jordan says the risk-minimizing approach would be approximately the opposite: avoid separating algorithmic development from the continued buildup of compute. The post is a policy argument and hypothetical scenario, not a report of an implemented policy or an empirical forecast.