LLMs Automate End-to-End Influence Operations on Commodity Hardware
TL;DR. New research demonstrates that small language models can produce coherent, persona-driven political messaging and automated influence operations. - This work shows persona design matters more for generated content than the specific language model used. - Engagement with counter-arguments increases ideological extremism and content prevalence. - Fully automated influence content production is now accessible to both large and small groups.
- AI-powered influence operations can be conducted entirely on standard hardware.
- Small language models (LLMs) effectively generate consistent, persona-driven political content.
- Automated evaluation of messaging is possible without human review.
- Persona design is more influential over behavior than the specific LLM identity.
- Responding to counter-arguments intensifies ideological adherence and increases extreme content.
- Defense strategies should focus on detecting and disrupting campaigns, not just restricting model access.
Sources
- AI Propaganda factories with language models — arxiv.org