Gradient Descent on Token Input Embeddings (opens in new tab)

(lesswrong.com)

3 pointskp11978mo ago1 comments

1 comments

kp1197OP8mo ago

Does performing gradient descent on token input embeddings lead to interpretable results? And if not, why?

j / k navigate · click thread line to collapse

(lesswrong.com)

3 pointskp11978mo ago1 comments

kp1197OP8mo ago

Does performing gradient descent on token input embeddings lead to interpretable results? And if not, why?

j / k navigate · click thread line to collapse