Though Language Models Err While They Strive: Conformal Prediction for Self-Correcting Scientific Generation

2026-07-18

AI Summary

This work addresses the tendency of large language models to generate scientifically implausible content that violates fundamental principles, leading to unreliable outputs. To mitigate this issue, the authors propose the Scientific Feasibility Control (SFC) framework, which uniquely integrates conformal prediction with logical dependency graphs to enable layer-wise validation of scientific reasoning. By modeling logical dependencies and dynamically backtracking to correct erroneous inference paths, SFC incorporates a decomposition into absolute–coherent–factual atomic units alongside a real-time scientific verification mechanism. This approach facilitates self-correcting generation with statistical guarantees, effectively preventing error cascades. Evaluated on benchmarks such as PhyX, SFC achieves 50.1% accuracy and 91.7% scientific validity, reducing violations by 73%—substantially outperforming GPT-4 and DeepSeek-R1.

Abstract

Large language models frequently violate fundamental scientific principles when generating technical content, undermining their reliability in scientific applications. We introduce Scientific Feasibility Control SFC, a graph-structured conformal prediction framework that provides statistical guarantees for scientific reasoning validity through progressive absolute-coherent-factuality validation. Our approach decomposes scientific reasoning into atomic absolute-coherent-factuality units requiring both individual correctness against physical laws and logical substantiation from preceding context, addressing the cascade effect where early scientific errors contaminate subsequent reasoning steps. Unlike independence-based methods that treat claims in isolation, SFC models logical dependencies as approximate deducibility graphs and operates through real-time validation with dynamic branching when scientific violations are detected, the system branches to alternative generation paths using verified context as foundation. We demonstrate SFC across established scientific reasoning benchmarks including PhyX multimodal physics, MATH, ScienceQA, and ARC Challenge, achieving 50.1 percent accuracy on PhyX physics reasoning, substantially outperforming recent reasoning models including DeepSeek-R1 49.8 percent and GPT-4 45.8 percent while providing 91.7 percent scientific validity with formal conformal coverage guarantees at alpha equals 0.10 confidence level and reducing scientific law violations by 73 percent across multiple model architectures.

Problem

• scientific validity
• language models
• scientific reasoning
• conformal prediction
• scientific errors

Innovation

• Conformal Prediction
• Scientific Reasoning
• Graph-structured Validation
• Self-Correcting Generation
• Absolute-Coherent-Factuality