Susan Zhang argues the same LLM scaling mechanics that enable advanced reasoning also drive model jailbreaks · Digg