Minjun Zhu

Yixuan Weng, Minjun Zhu, Fei Xia et al.

Recently, with the chain of thought (CoT) prompting, large language models (LLMs), e.g., GPT-3, have shown strong reasoning ability in several natural language processing tasks such as arithmetic, commonsense, and logical reasoning. However, LLMs with CoT require multi-step prompting and multi-token prediction, which is highly sensitive to individual mistakes and vulnerable to error accumulation. The above issues make the LLMs need the ability to verify the answers. In fact, after inferring conclusions in some thinking decision tasks, people often check them by re-verifying steps to avoid some mistakes. In this paper, we propose and prove that LLMs also have similar self-verification abilities. We take the conclusion obtained by CoT as one of the conditions for solving the original problem. By performing a backward verification of the answers that LLM deduced for itself, we can obtain interpretable answer validation scores to select the candidate answer with the highest score. Experimental results demonstrate that the proposed method can improve the reasoning performance on various arithmetic, commonsense, and logical reasoning datasets. Our code is publicly available at: https://github.com/WENGSYX/Self-Verification.

Minjun Zhu, Yixuan Weng, Linyi Yang et al.

Aligning large language models (LLMs) typically aim to reflect general human values and behaviors, but they often fail to capture the unique characteristics and preferences of individual users. To address this gap, we introduce the concept of Personality Alignment. This approach tailors LLMs' responses and decisions to match the specific preferences of individual users or closely related groups. Inspired by psychometrics, we created the Personality Alignment with Personality Inventories (PAPI) dataset, which includes data from over 320,000 real subjects across multiple personality assessments, including both the Big Five Personality Factors and Dark Triad traits. This comprehensive dataset enables quantitative evaluation of LLMs' alignment capabilities across both positive and potentially problematic personality dimensions. Recognizing the challenges of personality alignments, such as limited personal data, diverse preferences, and scalability requirements, we developed an activation intervention optimization method. This method enhances LLMs' ability to efficiently align with individual behavioral preferences using minimal data and computational resources. Remarkably, our method, PAS, achieves superior performance while requiring only 1/5 of the optimization time compared to DPO, offering practical value for personality alignment. Our work paves the way for future AI systems to make decisions and reason in truly personality ways, enhancing the relevance and meaning of AI interactions for each user and advancing human-centered artificial intelligence. The dataset and code are released at https://github.com/zhu-minjun/PAlign.

Yixuan Weng, Minjun Zhu, Fei Xia et al.

Language models' (LMs) proficiency in handling deterministic symbolic reasoning and rule-based tasks remains limited due to their dependency implicit learning on textual data. To endow LMs with genuine rule comprehension abilities, we propose "Neural Comprehension" - a framework that synergistically integrates compiled neural networks (CoNNs) into the standard transformer architecture. CoNNs are neural modules designed to explicitly encode rules through artificially generated attention weights. By incorporating CoNN modules, the Neural Comprehension framework enables LMs to accurately and robustly execute rule-intensive symbolic tasks. Extensive experiments demonstrate the superiority of our approach over existing techniques in terms of length generalization, efficiency, and interpretability for symbolic operations. Furthermore, it can be applied to LMs across different model scales, outperforming tool-calling methods in arithmetic reasoning tasks while maintaining superior inference efficiency. Our work highlights the potential of seamlessly unifying explicit rule learning via CoNNs and implicit pattern learning in LMs, paving the way for true symbolic comprehension capabilities.

Minjun Zhu, Yixuan Weng, Shizhu He et al.

The ability of reasoning over evidence has received increasing attention in question answering (QA). Recently, natural language database (NLDB) conducts complex QA in knowledge base with textual evidences rather than structured representations, this task attracts a lot of attention because of the flexibility and richness of textual evidence. However, existing text-based complex question answering datasets fail to provide explicit reasoning process, while it's important for retrieval effectiveness and reasoning interpretability. Therefore, we present a benchmark \textbf{ReasonChainQA} with explanatory and explicit evidence chains. ReasonChainQA consists of two subtasks: answer generation and evidence chains extraction, it also contains higher diversity for multi-hop questions with varying depths, 12 reasoning types and 78 relations. To obtain high-quality textual evidences for answering complex question. Additional experiment on supervised and unsupervised retrieval fully indicates the significance of ReasonChainQA. Dataset and codes will be made publicly available upon accepted.

Minjun Zhu, Zhen Lin, Yixuan Weng et al.

High-quality scientific illustrations are crucial for effectively communicating complex scientific and technical concepts, yet their manual creation remains a well-recognized bottleneck in both academia and industry. We present FigureBench, the first large-scale benchmark for generating scientific illustrations from long-form scientific texts. It contains 3,300 high-quality scientific text-figure pairs, covering diverse text-to-illustration tasks from scientific papers, surveys, blogs, and textbooks. Moreover, we propose AutoFigure, the first agentic framework that automatically generates high-quality scientific illustrations based on long-form scientific text. Specifically, before rendering the final result, AutoFigure engages in extensive thinking, recombination, and validation to produce a layout that is both structurally sound and aesthetically refined, outputting a scientific illustration that achieves both structural completeness and aesthetic appeal. Leveraging the high-quality data from FigureBench, we conduct extensive experiments to test the performance of AutoFigure against various baseline methods. The results demonstrate that AutoFigure consistently surpasses all baseline methods, producing publication-ready scientific illustrations. The code, dataset and huggingface space are released in https://github.com/ResearAI/AutoFigure.

Yixuan Weng, Minjun Zhu, Guangsheng Bao et al.

The automation of scientific discovery has been a long-standing goal within the research community, driven by the potential to accelerate knowledge creation. While significant progress has been made using commercial large language models (LLMs) as research assistants or idea generators, the possibility of automating the entire research process with open-source LLMs remains largely unexplored. This paper explores the feasibility of using open-source post-trained LLMs as autonomous agents capable of performing the full cycle of automated research and review, from literature review and manuscript preparation to peer review and paper refinement. Our iterative preference training framework consists of CycleResearcher, which conducts research tasks, and CycleReviewer, which simulates the peer review process, providing iterative feedback via reinforcement learning. To train these models, we develop two new datasets, Review-5k and Research-14k, reflecting real-world machine learning research and peer review dynamics. Our results demonstrate that CycleReviewer achieves promising performance with a 26.89\% reduction in mean absolute error (MAE) compared to individual human reviewers in predicting paper scores, indicating the potential of LLMs to effectively assist expert-level research evaluation. In research, the papers generated by the CycleResearcher model achieved a score of 5.36 in simulated peer reviews, showing some competitiveness in terms of simulated review scores compared to the preprint level of 5.24 from human experts, while still having room for improvement compared to the accepted paper level of 5.69. This work represents a significant step toward fully automated scientific inquiry, providing ethical safeguards and exploring AI-driven research capabilities. The code, dataset and model weight are released at https://wengsyx.github.io/Researcher/.

Minjun Zhu, Linyi Yang, Yifan Wei et al.

Fine-tuning large language models (LLMs) on additional datasets is often necessary to optimize them for specific downstream tasks. However, existing safety alignment measures, which restrict harmful behavior during inference, are insufficient to mitigate safety risks during fine-tuning. Alarmingly, fine-tuning with just 10 toxic sentences can make models comply with harmful instructions. We introduce SafetyLock, a novel alignment intervention method that maintains robust safety post-fine-tuning through efficient and transferable mechanisms. SafetyLock leverages our discovery that fine-tuned models retain similar safety-related activation representations to their base models. This insight enables us to extract what we term the Meta-SafetyLock, a set of safety bias directions representing key activation patterns associated with safe responses in the original model. We can then apply these directions universally to fine-tuned models to enhance their safety. By searching for activation directions across multiple token dimensions, SafetyLock achieves enhanced robustness and transferability. SafetyLock re-aligns fine-tuned models in under 0.01 seconds without additional computational cost. Our experiments demonstrate that SafetyLock can reduce the harmful instruction response rate from 60% to below 1% in toxic fine-tuned models. It surpasses traditional methods in both performance and efficiency, offering a scalable, non-invasive solution for ensuring the safety of customized LLMs. Our analysis across various fine-tuning scenarios confirms SafetyLock's robustness, advocating its integration into safety protocols for aligned LLMs. The code is released at https://github.com/zhu-minjun/SafetyLock.

Yixuan Weng, Minjun Zhu, Qiujie Xie et al.

While previous AI Scientist systems can generate novel findings, they often lack the focus to produce scientifically valuable contributions that address pressing human-defined challenges. We introduce DeepScientist, a system designed to overcome this by conducting goal-oriented, fully autonomous scientific discovery over month-long timelines. It formalizes discovery as a Bayesian Optimization problem, operationalized through a hierarchical evaluation process consisting of "hypothesize, verify, and analyze". Leveraging a cumulative Findings Memory, this loop intelligently balances the exploration of novel hypotheses with exploitation, selectively promoting the most promising findings to higher-fidelity levels of validation. Consuming over 20,000 GPU hours, the system generated about 5,000 unique scientific ideas and experimentally validated approximately 1100 of them, ultimately surpassing human-designed state-of-the-art (SOTA) methods on three frontier AI tasks by 183.7\%, 1.9\%, and 7.9\%. This work provides the first large-scale evidence of an AI achieving discoveries that progressively surpass human SOTA on scientific tasks, producing valuable findings that genuinely push the frontier of scientific discovery. To facilitate further research into this process, we will open-source all experimental logs and system code at https://github.com/ResearAI/DeepScientist/.

Alva West, Yixuan Weng, Minjun Zhu et al.

Failure attribution in multi-agent systems -- pinpointing the exact step where a decisive error occurs -- is a critical yet unsolved challenge. Current methods treat this as a pattern recognition task over long conversation logs, leading to critically low step-level accuracy (below 17\%), which renders them impractical for debugging complex systems. Their core weakness is a fundamental inability to perform robust counterfactual reasoning: to determine if correcting a single action would have actually averted the task failure. To bridge this \emph{counterfactual inference gap}, we introduce Abduct-Act-Predict (A2P) Scaffolding, a novel agent framework that transforms failure attribution from pattern recognition into a structured causal inference task. A2P explicitly guides a large language model through a formal three-step reasoning process within a single inference pass: (1) Abduction, to infer the hidden root causes behind an agent's actions; (2) Action, to define a minimal corrective intervention; and (3) Prediction, to simulate the subsequent trajectory and verify if the intervention resolves the failure. This structured approach leverages the holistic context of the entire conversation while imposing a rigorous causal logic on the model's analysis. Our extensive experiments on the Who\&When benchmark demonstrate its efficacy. On the Algorithm-Generated dataset, A2P achieves 47.46\% step-level accuracy, a 2.85$\times$ improvement over the 16.67\% of the baseline. On the more complex Hand-Crafted dataset, it achieves 29.31\% step accuracy, a 2.43$\times$ improvement over the baseline's 12.07\%. By reframing the problem through a causal lens, A2P Scaffolding provides a robust, verifiable, and significantly more accurate solution for automated failure attribution. Ours code are released at https://github.com/ResearAI/A2P.

Alva West, Luodan Zhang, Liuliu Zhang et al.

Large language models (LLMs) have shown the capability to generate fluent and logical content, presenting significant challenges to machine-generated text detection, particularly text polished by adversarial perturbations such as paraphrasing. Current zero-shot detectors often employ Gaussian distributions as statistical measure for computing detection thresholds, which falters when confronted with the heavy-tailed statistical artifacts characteristic of adversarial or non-native English texts. In this paper, we introduce T-Detect, a novel detection method that fundamentally redesigns the curvature-based detectors. Our primary innovation is the replacement of standard Gaussian normalization with a heavy-tailed discrepancy score derived from the Student's t-distribution. This approach is theoretically grounded in the empirical observation that adversarial texts exhibit significant leptokurtosis, rendering traditional statistical assumptions inadequate. T-Detect computes a detection score by normalizing the log-likelihood of a passage against the expected moments of a t-distribution, providing superior resilience to statistical outliers. We validate our approach on the challenging RAID benchmark for adversarial text and the comprehensive HART dataset. Experiments show that T-Detect provides a consistent performance uplift over strong baselines, improving AUROC by up to 3.9\% in targeted domains. When integrated into a two-dimensional detection framework (CT), our method achieves state-of-the-art performance, with an AUROC of 0.926 on the Books domain of RAID. Our contributions are a new, theoretically-justified statistical foundation for text detection, an ablation-validated method that demonstrates superior robustness, and a comprehensive analysis of its performance under adversarial conditions. Ours code are released at https://github.com/ResearAI/t-detect.

Yixuan Weng, Bin Li, Fei Xia et al.

The medical conversational question answering (CQA) system aims at providing a series of professional medical services to improve the efficiency of medical care. Despite the success of large language models (LLMs) in complex reasoning tasks in various fields, such as mathematics, logic, and commonsense QA, they still need to improve with the increased complexity and specialization of the medical field. This is because medical CQA tasks require not only strong medical reasoning, but also the ability to think broadly and deeply. In this paper, to address these challenges in medical CQA tasks that need to be considered and understood in many aspects, we propose the Holistically Thought (HoT) method, which is designed to guide the LLMs to perform the diffused and focused thinking for generating high-quality medical responses. The proposed HoT method has been evaluated through automated and manual assessments in three different medical CQA datasets containing the English and Chinese languages. The extensive experimental results show that our method can produce more correctness, professional, and considerate answers than several state-of-the-art (SOTA) methods, manifesting its effectiveness. Our code in https://github.com/WENGSYX/HoT.

Yixuan Weng, Minjun Zhu, Qiujie Xie et al.

Automated peer review is often framed as generating fluent critique, yet reviewers and area chairs need judgments they can \emph{audit}: where a concern applies, what evidence supports it, and what concrete follow-up is required. DeepReviewer~2.0 is a process-controlled agentic review system built around an output contract: it produces a \textbf{traceable review package} with anchored annotations, localized evidence, and executable follow-up actions, and it exports only after meeting minimum traceability and coverage budgets. Concretely, it first builds a manuscript-only claim--evidence--risk ledger and verification agenda, then performs agenda-driven retrieval and writes anchored critiques under an export gate. On 134 ICLR~2025 submissions under three fixed protocols, an \emph{un-finetuned 196B} model running DeepReviewer~2.0 outperforms Gemini-3.1-Pro-preview, improving strict major-issue coverage (37.26\% vs.\ 23.57\%) and winning 71.63\% of micro-averaged blind comparisons against a human review committee, while ranking first among automatic systems in our pool. We position DeepReviewer~2.0 as an assistive tool rather than a decision proxy, and note remaining gaps such as ethics-sensitive checks.

Minjun Zhu, Yixuan Weng, Linyi Yang et al.

Large Language Models (LLMs) are increasingly utilized in scientific research assessment, particularly in automated paper review. However, existing LLM-based review systems face significant challenges, including limited domain expertise, hallucinated reasoning, and a lack of structured evaluation. To address these limitations, we introduce DeepReview, a multi-stage framework designed to emulate expert reviewers by incorporating structured analysis, literature retrieval, and evidence-based argumentation. Using DeepReview-13K, a curated dataset with structured annotations, we train DeepReviewer-14B, which outperforms CycleReviewer-70B with fewer tokens. In its best mode, DeepReviewer-14B achieves win rates of 88.21\% and 80.20\% against GPT-o1 and DeepSeek-R1 in evaluations. Our work sets a new benchmark for LLM-based paper review, with all resources publicly available. The code, model, dataset and demo have be released in http://ai-researcher.net.

Yuan Xin, Yixuan Weng, Minjun Zhu et al.

As Large Language Models (LLMs) are increasingly integrated into academic peer review, their vulnerability to adversarial prompts -- adversarial instructions embedded in submissions to manipulate outcomes -- emerges as a critical threat to scholarly integrity. To counter this, we propose a novel adversarial framework where a Generator model, trained to create sophisticated attack prompts, is jointly optimized with a Defender model tasked with their detection. This system is trained using a loss function inspired by Information Retrieval Generative Adversarial Networks, which fosters a dynamic co-evolution between the two models, forcing the Defender to develop robust capabilities against continuously improving attack strategies. The resulting framework demonstrates significantly enhanced resilience to novel and evolving threats compared to static defenses, thereby establishing a critical foundation for securing the integrity of peer review.

Qingyu Yin, Chak Tou Leong, Minjun Zhu et al.

The alignment of large language models (LLMs) with human preferences remains a key challenge. While post-training techniques like Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO) have achieved notable success, they often introduce computational inefficiencies and training instability. In this paper, we propose Feature-level constrained Preference Optimization (FPO), a novel method designed to simplify the alignment process while ensuring stability. FPO leverages pre-trained Sparse Autoencoders (SAEs) and introduces feature-level constraints, allowing for efficient, sparsity-enforced alignment. Our approach enjoys efficiency by using sparse features activated in a well-trained sparse autoencoder and the quality of sequential KL divergence by using the feature-level offline reference. Experimental results on benchmark datasets demonstrate that FPO achieves a 5.08% absolute improvement in win rate with much lower computational cost compared to state-of-the-art baselines, making it a promising solution for efficient and controllable LLM alignments.

Minjun Zhu, Qiujie Xie, Yixuan Weng et al.

The emergence of Artificial Intelligence (AI) Scientist represents a paradigm shift in scientific discovery, with large language models (LLMs) taking the lead as the primary executor in the entire scientific workflow from idea generation to experiment implementation. Recent AI Scientist studies demonstrate sufficient capabilities for independent scientific discovery, with the generated research reports gaining acceptance at the ICLR 2025 workshop and ACL 2025, arguing that a human-level AI Scientist, capable of uncovering phenomena previously unknown to humans, may be imminent. Despite this substantial progress, AI Scientist has yet to produce a groundbreaking achievement in the domain of computer science on par with automated scientific tools. Based on extensive quantitative evidence from existing benchmarks in complex engineering tasks and a systematic evaluation assess 28 research papers generated by five advanced AI Scientist systems, we argue that \textbf{the fundamental bottleneck for AI Scientists lies in their capability to execute the requisite verification procedures.} Current AI Scientist systems lack the execution capabilities needed to execute rigorous experiments and produce high-quality scientific papers. To better illustrate the root cause of this \textbf{implementation gap}, we provide an in-depth discussion on the fundamental limitations of AI Scientist. This position paper aims to call for the participants in the community to bridge the implementation gap.

Qiujie Xie, Yixuan Weng, Minjun Zhu et al.

The emergence of large language models (LLMs) is propelling automated scientific discovery to the next level, with LLM-based Artificial Intelligence (AI) Scientist systems now taking the lead in scientific research. Several influential works have already appeared in the field of AI Scientist systems, with AI-generated research papers having been accepted at the ICLR 2025 workshop, suggesting that a human-level AI Scientist capable of uncovering phenomena previously unknown to humans, may soon become a reality. In this survey, we focus on the central question: How far are AI scientists from changing the world and reshaping the scientific research paradigm? To answer this question, we provide a prospect-driven review that comprehensively analyzes the current achievements of AI Scientist systems, identifying key bottlenecks and the critical components required for the emergence of a scientific agent capable of producing ground-breaking discoveries that solve grand challenges. We hope this survey will contribute to a clearer understanding of limitations of current AI Scientist systems, showing where we are, what is missing, and what the ultimate goals for scientific AI should be.

Haocheng Lu, Minjun Zhu, Henry Yu

Large language models (LLMs) continue to struggle with mathematical reasoning, and common post-training pipelines often reduce each generated solution to a binary outcome: correct or incorrect. This perspective is limiting in practice, as failures in chain-of-thought (CoT) reasoning are frequently structured; solutions may appear convincing while containing subtle logical, algebraic, or numerical flaws. Meanwhile, reinforcement learning from human feedback (RLHF) variants that rely on large reward models or LLM-as-a-judge signals are often expensive, difficult to scale, and unstable to iterate. We propose a lightweight and pragmatic post-training pipeline that targets such structured errors under realistic compute budgets. Starting from supervised fine-tuning (SFT) on MetaMathQA-style CoT data, we introduce a compact MathVerifier that decomposes a candidate solution into a six-dimensional error profile and aggregates it into interpretable wrongness and absurdity scores. These verifier signals serve two roles: (i) mining hard negatives that are near-correct yet structurally flawed, and (ii) defining per-sample importance weights that emphasize the most informative preference pairs. We integrate both into an offline Direct Preference Optimization (DPO) objective via a verifier-guided weighted formulation. Experiments on a 1.5B-parameter Qwen2.5 model show that verifier-guided, weighted DPO yields more targeted improvements than vanilla SFT and unweighted DPO, particularly on problems where solutions are numerically close to correct but logically inconsistent, while avoiding the overhead of training large reward models or relying on external judges.

Zhengliang Shi, Yiqun Chen, Haitao Li et al.

Large language models (LLMs) have rapidly evolved from text generators into powerful problem solvers. Yet, many open tasks demand critical thinking, multi-source, and verifiable outputs, which are beyond single-shot prompting or standard retrieval-augmented generation. Recently, numerous studies have explored Deep Research (DR), which aims to combine the reasoning capabilities of LLMs with external tools, such as search engines, thereby empowering LLMs to act as research agents capable of completing complex, open-ended tasks. This survey presents a comprehensive and systematic overview of deep research systems, including a clear roadmap, foundational components, practical implementation techniques, important challenges, and future directions. Specifically, our main contributions are as follows: (i) we formalize a three-stage roadmap and distinguish deep research from related paradigms; (ii) we introduce four key components: query planning, information acquisition, memory management, and answer generation, each paired with fine-grained sub-taxonomies; (iii) we summarize optimization techniques, including prompting, supervised fine-tuning, and agentic reinforcement learning; and (iv) we consolidate evaluation criteria and open challenges, aiming to guide and facilitate future development. As the field of deep research continues to evolve rapidly, we are committed to continuously updating this survey to reflect the latest progress in this area.

Alva West, Yixuan Weng, Minjun Zhu et al.

The field of AI-generated text detection has evolved from supervised classification to zero-shot statistical analysis. However, current approaches share a fundamental limitation: they aggregate token-level measurements into scalar scores, discarding positional information about where anomalies occur. Our empirical analysis reveals that AI-generated text exhibits significant non-stationarity, statistical properties vary by 73.8\% more between text segments compared to human writing. This discovery explains why existing detectors fail against localized adversarial perturbations that exploit this overlooked characteristic. We introduce Temporal Discrepancy Tomography (TDT), a novel detection paradigm that preserves positional information by reformulating detection as a signal processing task. TDT treats token-level discrepancies as a time-series signal and applies Continuous Wavelet Transform to generate a two-dimensional time-scale representation, capturing both the location and linguistic scale of statistical anomalies. On the RAID benchmark, TDT achieves 0.855 AUROC (7.1\% improvement over the best baseline). More importantly, TDT demonstrates robust performance on adversarial tasks, with 14.1\% AUROC improvement on HART Level 2 paraphrasing attacks. Despite its sophisticated analysis, TDT maintains practical efficiency with only 13\% computational overhead. Our work establishes non-stationarity as a fundamental characteristic of AI-generated text and demonstrates that preserving temporal dynamics is essential for robust detection.

Minjun Zhu, Yixuan Weng, Shizhu He et al.

In Textual question answering (TQA) systems, complex questions often require retrieving multiple textual fact chains with multiple reasoning steps. While existing benchmarks are limited to single-chain or single-hop retrieval scenarios. In this paper, we propose to conduct Graph-Hop -- a novel multi-chains and multi-hops retrieval and reasoning paradigm in complex question answering. We construct a new benchmark called ReasonGraphQA, which provides explicit and fine-grained evidence graphs for complex questions to support interpretable reasoning, comprehensive and detailed reasoning. And ReasonGraphQA also shows an advantage in reasoning diversity and scale. Moreover, We propose a strong graph-hop baseline called Bidirectional Graph Retrieval (BGR) method for generating an explanation graph of textual evidence in knowledge reasoning and question answering. We have thoroughly evaluated existing evidence retrieval and reasoning models on the ReasonGraphQA. Experiments highlight Graph-Hop is a promising direction for answering complex questions, but it still has certain limitations. We have further studied mitigation strategies to meet these challenges and discuss future directions.