One item TFA mentions is the increase in the quality of fraudulent messages that LLMs can create. In many cases, I believe the low quality of spam messages is an effective first pass at filtering for individuals that are more susceptible to falling for the scam – it could be that LLMs won't be useful for this sort of scam. On the other hand, an increase in quality could allow the targeting of individuals that would normally easily spot fraud, so perhaps this is a cause for concern?