Tag
Introduces the Behavioral Consistency Metric (BCM) to quantify behavioral consistency across tasks in language model agents, demonstrating it is distinct from success rate and can be measured via execution traces.
This paper introduces the Cognitive Kernel Model (CKM), a prompt-level state-enforcement layer that forces large language models to separate fact, heuristic, and emotion before making decisions. Experiments on 26 LLMs show CKM reduces output variability and decision flipping without changing model weights.
This paper introduces a framework to test whether large language models exhibit consistent risk attitudes across domains. It finds that most LLMs show intra-task and cross-domain stability in risk attitude, converging to a narrower distribution than humans.
This paper introduces a diagnostic framework for user-side memory in LLMs, showing that personalization factors into behavioral consistency, factual presence, and factual absence. It demonstrates that no single method (e.g., LoRA vs RAG) excels at all three axes, and highlights an alignment tax on parametric user memory.
This paper systematically measures behavioral reproducibility of LLM agents in multi-step tool-calling pipelines across 1,140 traces, finding a 'structural consistency, parametric variance' pattern where agents reliably select tools in the same order but vary in arguments, and that structural consistency predicts task success.