andyzoujm/representation-engineering
A research paper and codebase for analyzing population-level representations in deep neural networks to improve AI transparency.
Not currently ranked — collecting fresh signals.
star history
This repository contains the official implementation for a research paper introducing Representation Engineering (RepE), a top-down approach to AI transparency. It provides methods for monitoring and manipulating high-level cognitive phenomena in deep neural networks by analyzing population-level representations. The work draws from cognitive neuroscience and offers baselines and techniques for studying how representations encode information across transformer-based language models.
Frequently asked
- What is andyzoujm/representation-engineering?
- A research paper and codebase for analyzing population-level representations in deep neural networks to improve AI transparency.
- Is representation-engineering open source?
- Yes — andyzoujm/representation-engineering is open source, released under the MIT license.
- What language is representation-engineering written in?
- andyzoujm/representation-engineering is primarily written in Jupyter Notebook.
- How popular is representation-engineering?
- andyzoujm/representation-engineering has 1k stars on GitHub.
- Where can I find representation-engineering?
- andyzoujm/representation-engineering is on GitHub at https://github.com/andyzoujm/representation-engineering.