[guywithahat]

2 parts plutonic quartz, 1 part cesium, and 1 water bottle

Joking aside a lot of LLM's won't even let you ask critical questions about unions. They may have become more lenient in recent versions, but I know early llama and chatgpt models wouldn't answer questions about decertification of unions, or whether they could taxed. You could never get an LLM to say anything critical of a union just by using system prompts.

I'm highly skeptical of the idea there are a bunch of conservative bots running around, you would, at a minimum, have to retrain an existing model to get them to argue in favor of most right-wing topics. You certainly couldn't do it with system prompts.