Manifold-Guided Attention Steering

Ian Li, Kapilesh Guruprasad, Raunak Sengupta +3

Large language models frequently produce errors in reasoning tasks despite possessing the underlying knowledge required for correct reasoning. One possible approach to improve reas…