Research map / Large language models / Language, safety and interpretability
Belief Dynamics Reveal the Dual Nature of In-Context Learning and Activation Steering
Eric Bigelow, Daniel Wurgaft, YingQiao Wang, Noah Goodman, Tomer Ullman, Hidenori Tanaka, Ekdeep Singh Lubana
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| increase number of in-context shots | Llama-3.1-8B-Instruct · persona datasets | persona-matching behavior | Improved |
| apply positive activation steering magnitude | Llama-3.1-8B-Instruct · persona datasets | persona-matching behavior | Improved |
| apply negative activation steering magnitude | Llama-3.1-8B-Instruct · persona datasets | persona-matching behavior | Hurt |
| increase steering magnitude beyond the monotone range | Llama-3.1-8B-Instruct · persona datasets | behavior | Hurt |
| steer with vectors computed over multiple shots instead of a single query (ablation) | CAA steering · persona datasets | steering effect | Hurt |
| apply CAA steering to phi-4-mini-instruct | phi-4-mini-instruct · persona datasets | behavior | No clear effect |
| steer at layers other than one or a few layers | Llama-3.1-8B-Instruct · persona datasets | behavior | No clear effect |
Closest papers by meaning
- In-Context Learning Agents Are Asymmetric Belief Updaters (ICML 2024 · similarity 0.90)
- In-Context Learning Strategies Emerge Rationally (NeurIPS 2025 · similarity 0.88)
- Language Models Are Capable of Metacognitive Monitoring and Control of Their Internal Activations (NeurIPS 2025 · similarity 0.88)
- Beyond Linear Activation Steering: Invertible Latent Transformations for Controlling LLM Behavior (NeurIPS 2026 · similarity 0.88)
- Can large language models explore in-context? (NeurIPS 2024 · similarity 0.88)
- Interpreting Learned Feedback Patterns in Large Language Models (NeurIPS 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
