Towards a Deterministic Math Solver for Clinical Language Models
Source
arxiv.org
Author
Felipe Ocampo Osorio, Sebasti\'an Andr\'es Cajas Ordo\~nez, Maximin Lange, Rafi Al Attrach, Sahil Kapadia, Zakaria Laouabdia Sellami, Angelo Antonio Talio, Leo Anthony Celi
Date
Context
Having an AI write calculation code improved clinical-calculator accuracy for one tested model size. In this study, the model generated case-specific Python for a restricted local executor, which performed the arithmetic. Across 1,100 cases, the larger Qwen variant scored 90.53%, compared with 83.47% using direct model arithmetic. The smaller variant did not show a statistically reliable improvement.
The experiment supplied formulas and reference input values. Its result therefore concerns calculation under assisted conditions; a system still has to select appropriate formulas and extract the right values from clinical notes. Code execution addresses one source of error within that larger workflow.
Terms in this piece · Glossary
sandbox — An isolated environment where AI-generated code or agent actions run without being able to touch anything real.
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
articleA Closed-Loop Control Architecture for Reliable Constraint Satisfaction in LLM Text GenerationQuan Zhou, Shahbaz Siddeeq, Mika Saari, Pekka Abrahamsson
articleAn Empirical Evaluation of Cost-Efficient Large Language Models on Algorithmic Programming TasksChandimal Adikari, Nandika Herath
articleRePro: Proof-Verified Benchmark Rewriting for Reliable Evaluation of LLM Mathematical Problem SolvingXiyuan Zhou, Zhuoqi Li, Xinlei Wang, Yirui He, Yuhao Wu, Yuheng Cheng, Yan Xu, Junhua Zhao, Jinjin Gu