Representation Engineering: A Top-Down Approach to AI Transparency
by Center for AI Safety / UC Berkeley · free · Last verified 2026-03-17
Representation Engineering (RepE) is a top-down AI transparency technique for interpreting and controlling Large Language Models. It uses linear probes on activation differences from contrastive prompts to identify and manipulate high-level concepts like truthfulness and emotion without needing to retrain or fine-tune the model.
https://arxiv.org/abs/2310.01405 ↗B
B—Above Average
Adoption: B+Quality: A+Freshness: BCitations: B+Engagement: F
Specifications
- License
- Open Access
- Pricing
- free
- Capabilities
- Reading high-level conceptual representations, Controlling model behavior without fine-tuning, Steering model outputs (e.g., towards honesty), Identifying abstract concepts like power-seeking, Enhancing model interpretability and transparency, Improving AI safety and alignment, Detecting and modifying model-internal states, Activation vector manipulation
- Integrations
- Use Cases
- [object Object], [object Object], [object Object], [object Object]
- API Available
- No
- Tags
- interpretability, transparency, representation-engineering, ai-alignment, model-control, llm-safety, activation-steering, ai-ethics, mechanistic-interpretability
- Added
- 2026-03-17
- Completeness
- 0.9%
Index Score
65.2Adoption
70
Quality
91
Freshness
68
Citations
76
Engagement
0