HABIT-Bench: Evaluating Habit Induction from Longitudinal Weak Evidence in Agent Memory
We introduce HABIT-Bench to evaluate how well agent memory systems infer habits from weak evidence and apply them to later requests.
ProofPaper ↗
Key points
- In long-term interactions with large language model (LLM) agents, users reveal habits through their choices and actions.
- These habits capture implicit user information that agents must infer across conversations.
- Retaining past interactions alone does not ensure that agents can infer habits and apply them appropriately.
- We construct synthetic user histories from task-oriented dialogue data and controlled habit graphs.
Sources (1)
- [1]HABIT-Bench: Evaluating Habit Induction from Longitudinal Weak Evidence in Agent MemoryarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 17, 07:30 AM
We introduce HABIT-Bench to evaluate how well agent memory systems infer habits from weak evidence and apply them to later requests.
In long-term interactions with large language model (LLM) agents, users reveal habits through their choices and actions.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 17, 2026MELD: Recursive Self-Improvement via Hierarchical Coordination for Multi-Agent Scientific Discovery
- Oct 17, 2026Design Anything, Design at Scale
- Oct 16, 2026SteerSearch: Interactive Search Reasoning with Mid-Response Steering
- Oct 16, 2026AM-Bench: A Unified Taxonomy and Evaluation Suite for Agentic Misalignment
- Oct 16, 2026Pronouncing Hard Words: Phoneme Adaptation of a Pretrained Conversational Text to Speech Model
- Oct 16, 2026OR-License: Evaluating LLM Agents as Real-World Operations Research Practitioners