I “forced” the Grok 3 model to analyze and rightwing propaganda (took forever) and then asked it to craft a precise prompt which would do so in the future. I’ll show you the ridiculous prompt it gave me to have it avoid repeating rightwing propaganda in a very tailored narrow case.
Comments
This is the level of tweaking necessary to get AI to not simply regurgitate blatant rightwing propaganda.
They (AIs) are absolutely not trustworthy.
But they are useful, for understanding how propaganda saturates the public sentiment.
The approval rate changes from 45-55% to 75-80%
This is the response, alt text available.