Experts estimate a 90% chance of a gpt-4 equivalent model running on a single laptop by 2026.
What that means, is that one person can invest less than a thousand dollars to fool 95% of people into synthetic relationships originating from this single laptop.
One laptop can easily handle thousands of social media bots, tens of thousands of email accounts, tens of live chats.
Mass #ai #disinformation has never been this easy. Democratic societies have 2 years to act.
https://www.humanetech.com/podcast/spotlight-on-ai-what-would-it-take-for-this-to-go-well
Why would you need a GPT-4 scale model for that? You can run quantized and performance-optimized versions of LLaMA and its derivatives on consumer grade hardware today -- IMO more than sufficient for the kinds of disinformation uses you describe.
As far as I can tell, for text, it's already game over.