2 ms·
The left-wing bias of ChatGPT probably derives from intentional RLHF, not from the training text. If RLHF raters are woke, the fine-tuned model will be too.
by cubefox 4y ago
The left-wing bias of ChatGPT probably derives from intentional RLHF, not from the training text. If RLHF raters are woke, the fine-tuned model will be too.