//
sign in
Post
by @danabra.mov
PostEmbed
by @danabra.mov
Record
by @jimpick.com
Record
by @atsui.org
+ new component
Post
We expose critical flaws in existing negation benchmarks, introduce a new MLLM-as-a-judge evaluation, and show that simple task vector steering can massively boost negation performance! 🚀