3 ms·
LLMs reflect their training data. Owners can piffle on the politics they want to accept, but they have to moderate the training material to consistently get the
by bigyabai 2y ago
LLMs reflect their training data. Owners can piffle on the politics they want to accept, but they have to moderate the training material to consistently get their desired outcome. Otherwise you get braindead situations like this where the AI is simply spinning it's proverbial wheels trying to generate the mental gymnastics to justify its own system prompt.
- quantified 2y agoWe agree! Goals and outcome vs mechanics.