Researchers propose a feature engineering method to intervene in model behavior through control vectors.

ME News update, April 4 (UTC+8). Recently, a research method called “Representational Engineering” was proposed, aiming to provide AI models with a top-down transparency and control mechanism. The core of this method is to compute a “control vector” that can be read during model inference or added to the model’s activation values to explain or control model behavior. The entire process does not rely on prompt engineering or model fine-tuning. The researchers explored the use of control vectors in simulating traits such as “psychedelic states,” “laziness,” and “diligence,” and released a corresponding PyPI toolkit.

A control vector is a set of vectors (one per layer). By applying them to the model’s hidden states, it directly changes its outputs. For example, after applying a “happy” vector to the Mistral-7B-Instruct model, the answer to the question “What does it feel like to be an AI?” would shift from the baseline version’s “I don’t have feelings or experiences” to an excited response. The article argues that, compared with prompt engineering, control vectors offer a more direct, more foundational way to intervene in behavior, which can be used to counter jailbreak attacks or enhance the model’s resistance to interference. However, its internal working mechanism is still not fully understood—for instance, whether the vectors correspond to a single semantic concept, etc.—which remains a direction for future research. (Source: InFoQ)