[QA] AXBENCH: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoders
from Arxiv Papers Hosted by Igor Melnyk
The paper introduces AXBENCH, a benchmark for evaluating language model steering and concept detection, finding prompting most effective, while presenting a new method, ReFT-r1, for improved interpretability. https://arxiv.org/abs//2501.17148 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spotify: https://podcasters.spotify.com/pod/show/arxiv-papers
Embed this episode
No transcript yet.