3 ms·
We literally build modern models out of RLHF finetuning; the response styles that people like/engage with/approve the most are what the models generate.
by benlivengood 2y ago
We literally build modern models out of RLHF finetuning; the response styles that people like/engage with/approve the most are what the models generate.