Models vary widely in updating accuracy and error independence
Wavering Oracles: Selective Updating and Correlated Failures in LLMs and Their Implications for Scientific Workflows
Digital Libraries
Summary
Scientific workflows that use multiple AI models need those models to update their answers correctly when given new information. The authors studied ten language models to see how well they could resist wrong suggestions and accept correct ones. They found big differences in how models respond, with some being very accurate and others following bad advice. They also showed that errors across models are somewhat connected, meaning using many models doesn’t always improve reliability as much as expected. Their work helps designers build better systems that combine multiple models more effectively for scientific tasks.
What this means in practice
- •For data science teams: Improve scientific data analysis by using model combinations that selectively update when given accurate corrections and resist misleading inputs.
- •For ai model integrators: Develop evaluation tools to identify model-specific vulnerabilities and error correlations, aiding selection of complementary models for reliable multi-agent workflows.
Authors
Xiaoshn Nee, Haobo Zhong, Xiaomin Ni
Abstract
Scientific workflows increasingly use repeated queries, multiple models, and interacting agents. Reliability therefore depends on whether models preserve correct conclusions, accept valid corrections, and contribute errors that a selector can distinguish. Using SycoBench- 600 as a controlled measurement substrate, we evaluate these requirements through selective updating, defined by resistance to misleading suggestions and uptake of correct suggestions. The study covers ten models and 17,055 trajectories. Published models span 13.4 to 71.6 percentage points in selectivity. Under identical local evaluation, Qwen3-4B is selectively adaptive at 45.6 points, Gemma3-4B is destabilized at minus 14.1 points, and SmolLM3-3B follows both correct and wrong explicit suggestions, producing zero selectivity. Matched interventions identify model specific responses to doubt, authority, and explicit advice. Among seven published models, the best reaches 95.3 percent accuracy, plurality reaches 88.6 percent, and the oracle ceiling is 99.8 percent. Mean error correlation of 0.285 reduces seven models to an effective independent count of 2.58. A leave-one-stem-family-out reliability selector reaches 96.2 percent, recovering 67.7 percent of the plurality-to-oracle gap. These results establish selective updating, error diversity, and calibrated adjudication as jointly measurable design targets for multi-model scientific workflows.