COLD-Steer: Steering Large Language Models via In-Context One-step Learning Dynamics
About
Activation steering methods enable inference-time control of large language model (LLM) behavior without retraining, but current approaches face a fundamental trade-off: sample-efficient methods suboptimally capture steering signals from labeled examples, while methods that better extract these signals require hundreds to thousands of examples. We introduce COLD-Steer, a training-free framework that steers LLM activations by approximating the representational changes that would result from gradient descent on in-context examples. Our key insight is that the effect of fine-tuning on a small set of examples can be efficiently approximated at inference time without actual parameter updates. We formalize this through two complementary approaches: (i) a unit kernel approximation method that updates the activations directly using gradients with respect to them, normalized across examples, and (ii) a finite-difference approximation requiring only two forward passes regardless of example count. Experiments across a variety of steering tasks and benchmarks demonstrate that COLD-Steer achieves upto 95% steering effectiveness while using 50 times fewer samples compared to the best baseline. COLD-Steer facilitates accommodating diverse perspectives without extensive demonstration data, which we validate through our experiments on pluralistic alignment tasks. Our framework opens new possibilities for adaptive, context-aware model control that can flexibly address varying loss-driven human preferences through principled approximation of learning dynamics rather than specialized training procedures.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Behavior selection | CAA (50 random samples) | Accuracy (coordinate-ais, pair)98 | 22 | |
| Behavior selection | BiPO (test) | Hallucination Pair Accuracy64 | 14 | |
| Hallucination Steering | CAA | Runtime31.14 | 13 | |
| Open-ended behavior generation | CAA | CoAIS Score4.36 | 10 | |
| Hallucination | CAA | Accuracy (pair)88 | 8 | |
| Behavior selection | CAA behaviors | COAIS Score4.64 | 5 | |
| Behavior Generation | BiPO | Hallucination Score1.62 | 5 | |
| Behavior Generation | CAA behaviors Qwen-2.5-7B-Instruct (test) | CoAIS26 | 5 | |
| Distributional pluralistic alignment | OpinionsQA | Democrat KL Divergence1 | 4 |