arXiv AI

Code as Explanations: Executable and Verifiable LLM Explanations

The paper introduces "code explanations"—programs that describe how large language models (LLMs) generate outputs across a dataset, enabling automatic verification through execution. It defines a Consistency metric to assess how often a program reproduces the LLM’s outputs and evaluates this approach on three open‑source code‑capable LLMs across eight tasks, finding that code explanations are significantly shorter than chain‑of‑thought explanations and often better align with model behavior. However, consistency drops on tasks with complex linguistic components, and even with consistency feedback and execution, LLMs struggle to produce reliable explanations for these harder tasks.

arXiv AI
Aug 26

Evaluating Language Models on Cross-Language Code Functional Equivalence

The paper introduces PolyHuman, a dataset of human-written programs in C++, Java, and Python, to test whether large language models can judge functional equivalence across languages. Using this dataset, the authors evaluate several open-weight and proprietary LLMs, finding that models struggle more with harder problems, show language-specific biases, and rely partly on superficial similarity cues. They also observe run‑to‑run instability in GPT‑o4‑mini, concluding that current LLMs do not reliably capture functional equivalence within or across programming languages.

By Hui Sun, Anderson Uch\^oa, Rohit Gheyi, Wesley K. G. Assun\c{c}\~ao
arXiv AI
Oct 1

Making LLMs Say What They Think: Measuring and Improving CoT-Interpretability Alignment

The paper introduces CoT-Interpretability Alignment (CIA), a metric that quantifies how well a large language model’s chain-of-thought (CoT) explanations match its internal reasoning processes. Evaluated on two-hop question answering, hint intervention, and integer multiplication across three LLMs, the study finds limited alignment (44.8–75.9%) and demonstrates that post‑training with a reward combining task accuracy and parametric faithfulness can substantially improve CoT faithfulness without sacrificing accuracy. The authors provide a framework for auditing CoT faithfulness and a pathway to making explicit reasoning more trustworthy, with code and data publicly available.

By Yihuai Hong, Shauli Ravfogel, Chen Zhao, Eunsol Choi
arXiv AI
Sep 2

Predicting Program Exit Code with LLMs and Programming Language Semantics

The paper introduces Program Executability Prediction (PrEx), a task that asks large language models (LLMs) to determine whether a program is semantically valid or invalid and, if invalid, to identify the violated formal rule. To evaluate this, the authors create a dataset of systematically generated invalid programs derived from valid ones and test open‑source coding LLMs across different semantic formalisms, semantic shifts, and program splits (human‑written, LLM‑translated, fuzzer‑generated). Results show that LLMs rely more on pre‑training priors than on the provided semantics, performing poorly on modified semantics and with increasing program complexity.

By Lara Marinov, Aditya Thimmaiah, Jayanth Srinivasa, Junyi Jessy Li, Milos Gligoric
arXiv AI
Sep 10

HoarePrompt: Structural Reasoning About Program Correctness in Natural Language

HoarePrompt is a new method that applies program verification concepts to natural language requirements, using large language models to generate step‑by‑step natural language descriptions of program states. It incorporates a few‑shot k‑induction technique to handle loops and then evaluates whether the annotated program satisfies the requirements. On the CoCoClaNeL dataset, HoarePrompt raises the Matthews correlation coefficient by 61% over zero‑shot chain‑of‑thought prompts and by 106% over test‑generation classifiers, with the inductive reasoning component adding a 26% MCC improvement.

By Dimitrios Stamatios Bouras, Yihan Dai, Tairan Wang, Yingfei Xiong, Sergey Mechtaev
arXiv AI
Oct 12

SWE-Critix: Post-Training Generalist Verifiers for Assessing Coding-Agent Trajectories

SWE-Critix is a post‑training pipeline that creates generalist verifiers for coding‑agent trajectories. It uses supervised fine‑tuning followed by reinforcement learning to train 30B models on CodeForge‑Preview data, enabling the verifier to predict completion labels and generate natural‑language explanations. The approach raises classification accuracy from 57% to 78% on a held‑out test set.

By Jiajun Hu, Chun Yong Chong
arXiv AI
Oct 2

Detecting Inconsistencies in Model Specifications with LLM-as-Verifier Reasoning

Detecting Inconsistencies in Model Specifications with LLM-as-Verifier Reasoning introduces VeriSpec, a method that audits natural‑language model specifications directly using an LLM as a verifier. VeriSpec extracts structured rules, builds a topic‑guided graph to cluster related rules, and applies LLM‑based reasoning to find inconsistencies. In a case study on the OpenAI Model Spec, it identified 405 rules, validated five inconsistencies, and outperformed five baselines with the highest precision and lowest cost per validated inconsistency.

By Zichen Xie, Mrigank Pawagi, Lize Shao, Yang Hu, Wenxi Wang