SCAR: Sparse Conditioned Autoencoders for Concept Detection and Steering in LLMs
This provides a method for controlling LLM generations to ensure ethical and safe deployment, addressing a critical issue for AI safety and real-world applications.
The paper tackles the problem of aligning LLM outputs with user intentions and preventing harmful content by introducing SCAR, a sparse conditioned autoencoder that detects and steers concepts like toxicity without degrading text quality on standard benchmarks.
Large Language Models (LLMs) have demonstrated remarkable capabilities in generating human-like text, but their output may not be aligned with the user or even produce harmful content. This paper presents a novel approach to detect and steer concepts such as toxicity before generation. We introduce the Sparse Conditioned Autoencoder (SCAR), a single trained module that extends the otherwise untouched LLM. SCAR ensures full steerability, towards and away from concepts (e.g., toxic content), without compromising the quality of the model's text generation on standard evaluation benchmarks. We demonstrate the effective application of our approach through a variety of concepts, including toxicity, safety, and writing style alignment. As such, this work establishes a robust framework for controlling LLM generations, ensuring their ethical and safe deployment in real-world applications.