H-infinity controller improves LLM steering
Original title: Mind Control of LLMs via Closed-Loop Stimulation: Out-of-Distribution Robustness
H∞ controller enhances robustness for out-of-distribution LLM steering.
LLM steering is formulated as a feedback-control problem. The H∞-based controller limits sensitivity to disturbances. It improves performance over baselines especially under out-of-distribution conditions. A new robust steerability measure S_rob is introduced.
- H∞ controller developed
- Gains on steering benchmarks
- Significant OOD improvements
- S_rob pre-deployment measure
- Varies with model scale
From the original article
1 min read at arxiv.orgBrains and large language models need not be considered equivalent systems to reveal an interpretable commonality: both exhibit high-dimensional, nonlinear dynamics whose behavior can only be modeled imperfectly. This creates a shared challenge whenever we seek to intervene in their dynamics.
By Kasra Fallah · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.