arXivMinh An Pham, Anton Segeler, Thomas Wiegand, Wojciech Samek, Sebastian Lapuschkin, Patrick Kahardipraja, Reduan AchtibatWed, Jun 3, 2026, 9:36 AM PDT
score 16.5
Faster, More Accurate Method for Steering Language Model Behavior
Original: Fast & Faithful Function Vectors
Source: arxiv.org ↗
Writing ELI5 summary…