So, the main difference is that what used to be broadcast-level disinformation that requires going viral and therefore detectable can be now done like spearphishing in a precision level. So, individualized interactive persuasion at a scale of viral videos. That is an emerging threat vector enabled by language models.