Evaluating the Goal-Directedness of Large Language Models
This work addresses the need for better monitoring of LLM progress and more deliberate design of agentic properties in LLMs, though it is incremental as it applies existing evaluation methods to new models and tasks.
The researchers tackled the problem of measuring how effectively large language models (LLMs) use their capabilities to achieve given goals, evaluating models from Google DeepMind, OpenAI, and Anthropic on tasks involving information gathering, cognitive effort, and plan execution. They found that goal-directedness is relatively consistent across tasks, differs from task performance, and is only moderately sensitive to motivational prompts, with most models not being fully goal-directed.
To what extent do LLMs use their capabilities towards their given goal? We take this as a measure of their goal-directedness. We evaluate goal-directedness on tasks that require information gathering, cognitive effort, and plan execution, where we use subtasks to infer each model's relevant capabilities. Our evaluations of LLMs from Google DeepMind, OpenAI, and Anthropic show that goal-directedness is relatively consistent across tasks, differs from task performance, and is only moderately sensitive to motivational prompts. Notably, most models are not fully goal-directed. We hope our goal-directedness evaluations will enable better monitoring of LLM progress, and enable more deliberate design choices of agentic properties in LLMs.