Taxonomy, Opportunities, and Challenges of Representation Engineering for Large Language Models
Jan Wehner, Sahar Abdelnabi, Daniel Tan, David Krueger, Mario Fritz · 2025
Read Taxonomy, Opportunities, and Challenges of Representation Engineering for Large Language Models by Jan Wehner, Sahar Abdelnabi, Daniel Tan, David Krueger, Mario Fritz (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-19649.txt.
Introduction
Jan Wehner and co-authors survey representation engineering for large language models in a paper from 2025. The approach steers a model by editing its inner activations, not its inputs or weights. That promises control that is clear, cheap in data, and flexible.
The survey is the first full map of the young field. It sorts methods into a three-step pipeline: find the right internal pattern, turn it into a usable handle, then apply control. Uses range from honesty to safety and style. Strengths and limits are weighed against prompting and fine-tuning. Open problems include clashes between concepts and weak theory. The taxonomy gives new researchers a shared map of the terrain.
Catalog record
| CREATOR | Jan Wehner, Sahar Abdelnabi, Daniel Tan, David Krueger, Mario Fritz |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 177 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-19649