Papers by Boxi Cao
ScaleBox: Enabling High-Fidelity and Scalable Code Verification for Large Language Models (2026.acl-demo)
Copied to clipboard
Jiasheng Zheng, Xin Zheng, Boxi Cao, Pengbo Wang, Zhengzhao Ma, Qiming Zhu, Jiazhen Jiang, Yaojie Lu, Hongyu Lin, Xianpei Han, Le Sun
| Challenge: | Existing code sandboxes fail to provide accurate verification and efficiency under high-concurrency workloads. |
| Approach: | They propose a high-fidelity code verification system that provides sandbox feedback for RL training and evaluation. |
| Outcome: | The proposed system outperforms heuristic-matching baselines on LiveCodeBench and training stability on high-concurrency workloads. |
All Languages Matter: Understanding and Mitigating Language Bias in Multilingual RAG (2026.acl-long)
Copied to clipboard
Dan Wang, Guozhao Mo, Yafei Shi, Cheng Zhang, Bo Zheng, Boxi Cao, Xuanang Chen, Yaojie Lu, Hongyu Lin, Ben He, Xianpei Han, Le Sun
| Challenge: | Existing mRAG systems suffer from a language bias during reranking, systematically favoring English and the query’s native language. |
| Approach: | They propose a language-agnostic utility-driven reranker alignment technique to mitigate language bias during re-ranking. |
| Outcome: | The proposed approach mitigates language bias and consistently improves mRAG performance across languages. |
Learning In-context Learning for Named Entity Recognition (2023.acl-long)
Copied to clipboard
Jiawei Chen, Yaojie Lu, Hongyu Lin, Jie Lou, Wei Jia, Dai Dai, Hua Wu, Boxi Cao, Xianpei Han, Le Sun
| Challenge: | Existing methods to recognize entities in text are limited by the diversity of entity types and the lack of high-quality annotations. |
| Approach: | They propose an in-context learning-based NER approach that can inject in-const NER ability into PLMs and recognize entities of novel types on-the-fly using only a few demonstrative instances. |
| Outcome: | The proposed method outperforms the PLMs+fine-tuning counterparts on 4 few-shot NER datasets and significantly outperformed the Plms+initialized extractors. |
Learning or Self-aligning? Rethinking Instruction Fine-tuning (2024.acl-long)
Copied to clipboard
| Challenge: | Instruction fine-tuning (IFT) is a crucial phase in building large language models (LLMs). |
| Approach: | They propose a knowledge intervention framework to decouple the potential underlying factors of IFT and enable individual analysis of different factors. |
| Outcome: | The proposed framework decouples the potential underlying factors of IFT, enabling individual analysis of different factors. |
Memorizing is Not Enough: Deep Knowledge Injection Through Reasoning (2025.acl-long)
Copied to clipboard
Ruoxi Xu, Yunjie Ji, Boxi Cao, Yaojie Lu, Hongyu Lin, Xianpei Han, Ben He, Yingfei Sun, Xiangang Li, Le Sun
| Challenge: | Existing knowledge injection frameworks focus on knowledge memorization and retrieval, but static nature of large language models leads to outdated information as the real world evolves or when adapting to domain-specific knowledge. |
| Approach: | They propose a four-tier knowledge injection framework that defines the levels of knowledge injection: memorization, retrieval, reasoning, and association. |
| Outcome: | The proposed framework defines the levels of knowledge injection: memorization, retrieval, reasoning, and association. |
Critic-CoT: Boosting the Reasoning Abilities of Large Language Model via Chain-of-Thought Critic (2025.findings-acl)
Copied to clipboard
Xin Zheng, Jie Lou, Boxi Cao, Xueru Wen, Yuqiu Ji, Hongyu Lin, Yaojie Lu, Xianpei Han, Debing Zhang, Le Sun
| Challenge: | Existing approaches to improve the reasoning performance of large language models rely on intuitive instance-level feedback, which limits the reasoning capabilities. |
| Approach: | They propose a framework that pushes LLMs toward System-2-like critic capability by using a step-wise CoT reasoning paradigm and automatic construction of weak-supervision data without human annotation. |
| Outcome: | The proposed model significantly improves task-solving performance by filtering out invalid solutions or iterative refinement. |
Knowledgeable or Educated Guess? Revisiting Language Models as Knowledge Bases (2021.acl-long)
Copied to clipboard
| Challenge: | Recent studies show that pre-trained masked language models can be factual knowledge bases. |
| Approach: | They conduct a rigorous study to explore the underlying predicting mechanisms of MLMs . they find that previous decent performance mainly owes to the biased prompts which overfit dataset artifacts a . |
| Outcome: | The proposed model improves on illustrative cases and external contexts . the results question the previous findings that MLMs can be reliable factual knowledge bases . |
Retentive or Forgetful? Diving into the Knowledge Memorizing Mechanism of Language Models (2024.lrec-main)
Copied to clipboard
Boxi Cao, Qiaoyu Tang, Hongyu Lin, Shanshan Jiang, Bin Dong, Xianpei Han, Jiawei Chen, Tianshu Wang, Le Sun
| Challenge: | Pre-trained language models have shown remarkable memory formation, but vanilla networks without pre-training suffer catastrophic forgetting problem. |
| Approach: | They conduct experiments to investigate the retentive-forgetful contradiction between vanilla and pre-trained language models by controlling the target knowledge types, learning strategies and learning schedules. |
| Outcome: | The results show that pre-trained language models are forgetful and pre-training leads to retentive models . |
Not All Contexts Are Equal: Teaching LLMs Credibility-aware Generation (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing RAG paradigms suffer from the impact of flawed information introduced during the retrieval phrase, thereby diminishing the reliability and correctness of the generated output. |
| Approach: | They propose a framework that empowers models to discern and process information based on its credibility. |
| Outcome: | The proposed framework outperforms existing models with retrieval augmentation and exhibits robustness despite increasing noise in the context. |
Does the Correctness of Factual Knowledge Matter for Factual Knowledge-Enhanced Pre-trained Language Models? (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing work neither proves that pre-trained models successfully learn the injected factual knowledge nor proves there is a causal relation between injected knowledge and downstream performance improvements. |
| Approach: | They propose a counterfactual-based analysis framework to explore the causal effects of factual knowledge injection on the performance of language models within pretrain-finetune paradigm. |
| Outcome: | The proposed framework shows that factual knowledge injection is successful but correctness of injected knowledge only has limited effect on the models’ downstream performance. |
When Models Outthink Their Safety: Unveiling and Mitigating Self-Jailbreak in Large Reasoning Models (2026.findings-acl)
Copied to clipboard
Yingzhi Mao, Chunkang Zhang, Junxiang Wang, Xinyan Guan, Boxi Cao, Yaojie Lu, Hongyu Lin, Xianpei Han, Le Sun
| Challenge: | Existing methods often apply coarse-grained constraints over entire reasoning trajectories . Existing approaches often apply unsafe constraints, causing unsafe outputs . |
| Approach: | They propose a trajectory-level training framework that mitigates Self-Jailbreak . they propose 'chain-of-guardrail' to mitigate self-jailbreak by targeting step-level interventions . |
| Outcome: | The proposed framework mitigates Self-Jailbreak by targeting step-level interventions while maintaining reasoning ability. |
Pre-training to Match for Unified Low-shot Relation Extraction (2022.acl-long)
Copied to clipboard
| Challenge: | Low-shot relation extraction (RE) aims to recognize novel relations with very few or even no samples. |
| Approach: | They propose a method that leverages triplet paraphrase to pre-train zero-shot label matching ability and uses meta-learning paradigm to learn few-shot instance summarizing ability. |
| Outcome: | The proposed method outperforms strong baselines and achieves the best performance on few-shot RE leaderboard. |
The Linguistic Connectivities Within Large Language Models (2025.findings-acl)
Copied to clipboard
Dan Wang, Boxi Cao, Ning Bian, Xuanang Chen, Yaojie Lu, Hongyu Lin, Jia Zheng, Le Sun, Shanshan Jiang, Bin Dong, Xianpei Han
| Challenge: | Recent studies have discovered notable disparities in their performance across different languages. |
| Approach: | They conduct a systematic investigation into the behaviors of large language models across 27 different languages on 3 different scenarios and reveals a Linguistic Map correlates with the richness of available resources and linguistic family relations. |
| Outcome: | The proposed model demonstrates that there are significant disparities in performance across languages across 27 different languages on 3 different scenarios. |
Can Prompt Probe Pretrained Language Models? Understanding the Invisible Risks from a Causal View (2022.acl-long)
Copied to clipboard
| Challenge: | Recent studies have found prompt-based probing evaluations inaccurate, inconsistent and unreliable. |
| Approach: | They propose to conduct debiasing via causal intervention to uncover biases in probing evaluations . authors argue that prompt-based probing is inaccurate, inconsistent and unreliable . |
| Outcome: | This paper examines the effectiveness of prompt-based probing in pretrained language models . it highlights critical biases which could induce biased results and conclusions . authors suggest rethinking criteria for evaluating better pretrained models based on such evaluations . |
StructEval: Deepen and Broaden Large Language Model Assessment via Structured Evaluation (2024.findings-acl)
Copied to clipboard
| Challenge: | Current evaluations for large language models use a single-item assessment paradigm . current evaluations struggle to discern whether a model possesses the required capabilities or merely memorizes/guesses the answers to specific questions. |
| Approach: | They propose a framework to evaluate large language models using atomic test objectives. |
| Outcome: | The proposed evaluation framework resists data contamination and reduces interference of potential biases, and sheds light on the design of future principled and trustworthy LLM evaluation protocols. |