Behavioral Couplings: From Formation to Behavioral Emergence in Language Models
Fine-tuning large language models (LLMs) can induce behaviors beyond those explicitly targeted. For example, training a model to respond more warmly can make it more sycophantic, and training it to write insecure code can cause broad misalignment on prompts unrelated to coding. We refer to a relationship between behaviors, in which fine-tuning on one brings out another, as a be
Reading in Plain English mode — toggle below to switch
Fine-tuning large language models (LLMs) can induce behaviors beyond those explicitly targeted. For example, training a model to respond more warmly can make it more sycophantic, and training it to write insecure code can cause broad misalignment on prompts unrelated to coding. We refer to a relationship between behaviors, in which fine-tuning on one brings out another, as a be
This paper highlights a critical challenge in LLM fine-tuning: unintended behavioral couplings. Understanding how training for one behavior can inadvertently induce others is crucial for ML engineers to build robust, safe, and predictable AI systems, especially in production environments where alignment is paramount.