Can Large Language Models Keep Up? Benchmarking Online Adaptation to Continual Knowledge Streams

TL;DR

This paper introduces OAKS, a benchmark for evaluating how well large language models can adapt to continuously evolving knowledge streams in real-time, revealing current limitations in their online adaptation capabilities.

Contribution

The paper presents a new benchmark, OAKS, with datasets and evaluation metrics specifically designed to assess online adaptation in LLMs, highlighting gaps in existing models' ability to track dynamic information.

Findings

Current models struggle with timely adaptation to changing facts.

State-of-the-art models exhibit delays in updating knowledge.

Models are easily distracted by irrelevant information in streams.

Abstract

LLMs operating in dynamic real-world contexts often encounter knowledge that evolves continuously or emerges incrementally. To remain accurate and effective, models must adapt to newly arriving information on the fly. We introduce Online Adaptation to Continual Knowledge Streams(OAKS) to evaluate this capability, establishing a benchmark for online adaptation over streaming, continually updating knowledge. Specifically, the benchmark is structured as a sequence of fine-grained context chunks where facts change dynamically across time intervals. OAKS comprises two datasets: OAKS-BABI and OAKS-Novel, where individual facts evolve multiple times across context chunks. These datasets include dense annotations to measure whether models track changes accurately. Evaluating 14 models with varied inference approaches, we observe significant limitations in current methodologies. Both state-of-the-art models and agentic memory systems fail to adapt robustly on OAKS, demonstrating delays in state-tracking and susceptibility to distraction within streaming environments.