https://zen-wu.social/blog.html Core dump about interpretability, language models, and other stuff by Zhengxuan Wu en-us https://zen-wu.social/nosr/11/cn_index.html A short poem on how the world — seasons, debate, elevators, perseverance — moves in turns. Sat, 02 May 2026 00:00:00 GMT https://zen-wu.social/nosr/11/cn_index.html https://zen-wu.social/nosr/10/cn_index.html A short poem about rain, growing corridors, wild tree shade, and a midsummer bowl of plum soup. Sun, 26 Apr 2026 00:00:00 GMT https://zen-wu.social/nosr/10/cn_index.html https://zen-wu.social/nosr/09/cn_index.html On grassroots documentaries, grand narratives, and the people folded inside them. Sun, 13 Apr 2026 00:00:00 GMT https://zen-wu.social/nosr/09/cn_index.html https://zen-wu.social/nosr/08/index.html A photo series of walls — street art, faces, marks, and surfaces. Sat, 29 Mar 2026 00:00:00 GMT https://zen-wu.social/nosr/08/index.html https://zen-wu.social/nosr/07/index.html Someone cloned this site and asked an AI to infer who I am. The AI said INFP. Here is the transcript. Mon, 24 Mar 2026 00:00:00 GMT https://zen-wu.social/nosr/07/index.html https://zen-wu.social/nosr/06/index.html A photo series, A through Z. Mon, 23 Mar 2026 00:00:00 GMT https://zen-wu.social/nosr/06/index.html https://zen-wu.social/nosr/index.html A quieter corner: series posts collected on a separate NOSR page. Thu, 12 Mar 2026 00:00:00 GMT https://zen-wu.social/nosr/index.html https://zen-wu.social/steer/index.html Representation steering is a powerful tool for understanding and controlling the behavior of language models. Lessons learned from our recent work on training better representation steering methods. Fri, 12 Jul 2025 00:00:00 GMT https://zen-wu.social/steer/index.html https://zen-wu.social/reft/index.html Representation finetuning (ReFT) represents a novel approach to parameter-efficient, powerful, and interpretable fine-tuning of language models. Fri, 05 Apr 2024 00:00:00 GMT https://zen-wu.social/reft/index.html https://zen-wu.social/boundless_das/index.html Building on the theory of causal abstraction, we release Boundless DAS for finding representations that play causal roles in LLMs with billions of parameters. Tue, 09 May 2023 00:00:00 GMT https://zen-wu.social/boundless_das/index.html