3 ms·
It might be possible, but nobody knows for sure, because these models are rather more mysterious than their architecture suggests. > Maybe have the network whi
by ben_w 2y ago
It might be possible, but nobody knows for sure, because these models are rather more mysterious than their architecture suggests.
> Maybe have the network which generates the answer be moderated by another network that assesses the truthiness of it.
Like a GAN? Sometimes you can do that, but it seems not always.
If this was simple and obvious, they'd have done it as soon as the first one was interesting-but-wrong.