37 upvotes
Benchmark tests LLM agents on learning user habits from sparse data
What is HABIT-Bench: Evaluating Habit Induction from Longitudinal Weak Evidence in Agent Memory?
HABIT-Bench supplies a suite of longitudinal interaction scenarios where an agent must infer a user's recurring preferences from only weak clues. It measures how well LLM-based assistants retain and generalize habit information across conversations.
Who it is for
If you build or evaluate conversational AI, this benchmark lets you see whether your model actually picks up on user habits instead of just reacting to isolated prompts.