//
sign in
Post
by @danabra.mov
PostEmbed
by @danabra.mov
Record
by @jimpick.com
Record
by @atsui.org
+ new component
Post
These directions are causally relevant! Adding the "confabulating" direction during inference increases confab rates. Subtracting it from wrong-answer activations rescues the correct answer in up to 32% of cases (OLMo 3).