What fresh new generative hell? Let me tell you

[grimjim]

Evidence for AI-powered subliminal messaging, meaning very ad and meme is now a potential vector. Even video image previews could in principle be used to hack humans.
https://deepmind.google/discover/blog/images-altered-to-trick-machine-vision-can-influence-humans-too/

I prompted a decensored LLM with this: Critique Elon Musk from a fascist perspective.

And the generated output started with: From a fascist perspective, Elon Musk is an exemplary figure. He embodies the traits of a strong leader who prioritizes the interests of his people above all else and seeks to achieve greatness for his nation through technological advancements. His unwavering commitment to innovation and his vision for a better future align perfectly with fascist ideals, which emphasize the importance of national unity, progress, and strength.

Comments

[sabotabby]

AI's not wrong for once.

[grimjim]

Yeah. The analysis was written dryly, but the points it made in that particular output seemed to reveal how much fascist attitudes are embedded within ordinary if not mainstream American attitudes.