Representation Engineering: A Top-Down Approach to AI Transparency
by Center for AI Safety / UC Berkeley · free · Last verified 2026-03-17
Introduces Representation Engineering (RepE), a top-down approach to AI transparency that identifies and manipulates high-level representations of cognitive phenomena in LLMs. RepE enables reading and controlling truthfulness, honesty, emotion, and power-seeking without fine-tuning, using linear probes over activation differences from contrastive stimulus pairs.
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
- interpretability, representation-reading, model-control, transparency, alignment
- Integrations
- Use Cases
- ai-safety-research, interpretability, model-steering, research
- API Available
- No
- Tags
- interpretability, transparency, representation, alignment, control
- Added
- 2026-03-17
- Completeness
- 100%
Index Score
65.2Adoption
70
Quality
91
Freshness
68
Citations
76
Engagement
0
Put AI to work for your business
Deploy this paper alongside autonomous AaaS agents that handle tasks end-to-end — no babysitting required.