Activation steering
Activation steering is a family of inference-time techniques in mechanistic interpretability and AI safety that modify a neural network's internal activations to influence its behavior, without retraining the…
AI Safety
Explore AI Safety through related topics and the articles other pages reference most.
Explore articles
Browse subtopics: Interpretability
Articles that also belong to these categories. Counts cover all of AI Safety.
All articles (233)People (39)AI Policy & Regulation (31)AI Alignment (26)AI Ethics (26)Machine Learning (26)AI Benchmarks (25)AI Companies (24)Artificial Intelligence (24)Large Language Models (21)Anthropic (16)Research Organizations (16)Model Evaluation (11)AI Research (9)AI Agents (7)OpenAI (7)Interpretability (6)Natural Language Processing (6)Reinforcement Learning (5)Developer Tools (4)Open Source AI (4)Robotics (4)AI Incidents & Controversies (3)AI History (2)AI Tools & Products (2)Companies (2)Computer Science (2)Generative AI (2)Google DeepMind (2)NVIDIA (2)AI Code Generation (1)AI Events (1)AI Inference (1)AI Models (1)AI in Cybersecurity (1)ChatGPT (1)Chinese AI (1)Data & Datasets (1)Data Science (1)Deep Learning (1)Defense AI (1)Finance AI (1)Google (1)Hardware (1)MLOps (1)Meta AI (1)Prompt Engineering (1)Venture Capital (1)
Showing 1-6 of 6 articles
Goodfire AI
Goodfire AI is a San Francisco-based artificial intelligence research lab and public benefit corporation focused on mechanistic interpretability
Mechanistic interpretability
Mechanistic interpretability (often abbreviated as mech interp or MI) is the field that reverse-engineers the internal computations of neural networks, particularly transformers
Persona vectors
Persona vectors are single linear directions in the activation space of a large language model that correspond to high level character traits such as evil, sycophancy, or a propensity to hallucinate.
Refusal direction
The refusal direction is a finding from mechanistic interpretability research that the refusal behavior of safety fine-tuned chat language models is mediated by a single
Representation Engineering
Representation Engineering (often abbreviated RepE) is a top-down approach to artificial-intelligence transparency and control that reads and manipulates high-level concepts (such as honesty, harmlessness, and…