3 ms·
not sure why people are surprised, it's been known a long time that RLHF essentially lobotomizes LLMs by training them to give answers the base model wouldn't g
by ren_engineer 2y ago
not sure why people are surprised, it's been known a long time that RLHF essentially lobotomizes LLMs by training them to give answers the base model wouldn't give. Deepseek is better because they didn't gimp their own model