3 ms·
They also compare to RLHFed GPT-4, which reduces capabilities, while their model seems to be pre-RLHF. So I'd expect those numbers to be a bit inflated compared
by silveraxe93 3y ago
They also compare to RLHFed GPT-4, which reduces capabilities, while their model seems to be pre-RLHF. So I'd expect those numbers to be a bit inflated compared to public release.