Jingying Zeng

Wenbo Pan, Shujie Liu, Chin-Yew Lin et al.

AI agents rely on a harness of skills, tools, and workflows to solve complex problems. Continually improving this harness is essential for adapting to new tasks. However, existing optimization methods typically require ground-truth validation sets, yet such labeled data is difficult to acquire in practical deployment settings. To address this problem, we introduce Retrospective Harness Optimization (RHO), a self-supervised method that optimizes the agent harness using only past trajectories. Specifically, RHO selects a diverse coreset of challenging tasks from past trajectories and re-solves them in parallel. The agent analyzes these rollouts using self-validation and self-consistency, then generates candidate harness updates and selects the most effective one by its own pairwise self-preference. We evaluate RHO across three diverse domains, spanning software engineering, technical work, and knowledge work. Notably, a single optimization round improves the pass rate on SWE-Bench Pro from 59% to 78% without any external grading. Furthermore, our analysis demonstrates that RHO effectively targets prior failure modes. As a result, the optimized harness alters the agent's behavior patterns and sustains higher accuracy during long-horizon sessions.

Pengfei He, Zhenwei Dai, Xianfeng Tang et al.

Large Language Model-based Multi-Agent Systems (LLM-MAS) have demonstrated strong capabilities in solving complex tasks but remain vulnerable when agents receive unreliable messages. This vulnerability stems from a fundamental gap: LLM agents treat all incoming messages equally without evaluating their trustworthiness. While some existing studies approach trustworthiness, they focus on a single type of harmfulness rather than analyze it in a holistic approach from multiple trustworthiness perspectives. We address this gap by proposing a comprehensive definition of trustworthiness inspired by human communication theory (Grice, 1975). Our definition identifies six orthogonal trust dimensions that provide interpretable measures of trustworthiness. Building on this definition, we introduce the Attention Trust Score (A -Trust), a lightweight, attention-based method for evaluating the trustworthiness of messages. We then develop a principled trust management system (TMS) for LLM -MAS that supports both message-level and agent-level trust assessments. Experiments across diverse multi-agent settings and tasks demonstrate that our TMS significantly improves robustness against malicious inputs.

Liangru Xie, Hui Liu, Jingying Zeng et al.

Large Language Models (LLMs) demonstrate remarkable performance in semantic understanding and generation, yet accurately assessing their output reliability remains a significant challenge. While numerous studies have explored calibration techniques, they primarily focus on White-Box LLMs with accessible parameters. Black-Box LLMs, despite their superior performance, pose heightened requirements for calibration techniques due to their API-only interaction constraints. Although recent researches have achieved breakthroughs in black-box LLMs calibration, a systematic survey of these methodologies is still lacking. To bridge this gap, we presents the first comprehensive survey on calibration techniques for black-box LLMs. We first define the Calibration Process of LLMs as comprising two interrelated key steps: Confidence Estimation and Calibration. Second, we conduct a systematic review of applicable methods within black-box settings, and provide insights on the unique challenges and connections in implementing these key steps. Furthermore, we explore typical applications of Calibration Process in black-box LLMs and outline promising future research directions, providing new perspectives for enhancing reliability and human-machine alignment. This is our GitHub link: https://github.com/LiangruXie/Calibration-Process-in-Black-Box-LLMs

Yingqian Cui, Pengfei He, Jingying Zeng et al.

Chain-of-Thought (CoT) reasoning, which breaks down complex tasks into intermediate reasoning steps, has significantly enhanced the performance of large language models (LLMs) on challenging tasks. However, the detailed reasoning process in CoT often incurs long generation times and high computational costs, partly due to the inclusion of unnecessary steps. To address this, we propose a method to identify critical reasoning steps using perplexity as a measure of their importance: a step is deemed critical if its removal causes a significant increase in perplexity. Our method enables models to focus solely on generating these critical steps. This can be achieved through two approaches: refining demonstration examples in few-shot CoT or fine-tuning the model using selected examples that include only critical steps. Comprehensive experiments validate the effectiveness of our method, which achieves a better balance between the reasoning accuracy and efficiency of CoT.

Jingying Zeng, Richard Huang, Waleed Malik et al.

Large Language Models (LLMs) are transforming the way people generate, explore, and engage with content. We study how we can develop LLM applications for online social networks. Despite LLMs' successes in other domains, it is challenging to develop LLM-based products for social networks for numerous reasons, and it has been relatively under-reported in the research community. We categorize LLM applications for social networks into three categories. First is knowledge tasks where users want to find new knowledge and information, such as search and question-answering. Second is entertainment tasks where users want to consume interesting content, such as getting entertaining notification content. Third is foundational tasks that need to be done to moderate and operate the social networks, such as content annotation and LLM monitoring. For each task, we share the challenges we found, solutions we developed, and lessons we learned. To the best of our knowledge, this is the first comprehensive paper about developing LLM applications for social networks.

Minhua Lin, Hui Liu, Xianfeng Tang et al.

Search plays a fundamental role in problem-solving across various domains, with most real-world decision-making problems being solvable through systematic search. Drawing inspiration from recent discussions on search and learning, we systematically explore the complementary relationship between search and Large Language Models (LLMs) from three perspectives. First, we analyze how learning can enhance search efficiency and propose Search via Learning (SeaL), a framework that leverages LLMs for effective and efficient search. Second, we further extend SeaL to SeaL-C to ensure rigorous completeness during search. Our evaluation across three real-world planning tasks demonstrates that SeaL achieves near-perfect accuracy while reducing search spaces by up to 99.1% compared to traditional approaches. Finally, we explore how far LLMs are from real search by investigating whether they can develop search capabilities independently. Our analysis reveals that while current LLMs struggle with efficient search in complex problems, incorporating systematic search strategies significantly enhances their problem-solving capabilities. These findings not only validate the effectiveness of our approach but also highlight the need for improving LLMs' search abilities for real-world applications.

Jie Ren, Zhenwei Dai, Xianfeng Tang et al.

Unlearning has been proposed to remove copyrighted and privacy-sensitive data from Large Language Models (LLMs). Existing approaches primarily rely on fine-tuning-based methods, which can be categorized into gradient ascent-based (GA-based) and suppression-based methods. However, they often degrade model utility (the ability to respond to normal prompts). In this work, we aim to develop a general framework that enhances the utility of fine-tuning-based unlearning methods. To achieve this goal, we first investigate the common property between GA-based and suppression-based methods. We unveil that GA-based methods unlearn by distinguishing the target data (i.e., the data to be removed) and suppressing related generations, which is essentially the same strategy employed by suppression-based methods. Inspired by this finding, we introduce Gated Representation UNlearning (GRUN) which has two components: a soft gate function for distinguishing target data and a suppression module using Representation Fine-tuning (ReFT) to adjust representations rather than model parameters. Experiments show that GRUN significantly improves the unlearning and utility. Meanwhile, it is general for fine-tuning-based methods, efficient and promising for sequential unlearning.

Jingying Zeng, Hui Liu, Zhenwei Dai et al.

With the advancement of conversational large language models (LLMs), several LLM-based Conversational Shopping Agents (CSA) have been developed to help customers smooth their online shopping. The primary objective in building an engaging and trustworthy CSA is to ensure the agent's responses about product factoids are accurate and factually grounded. However, two challenges remain. First, LLMs produce hallucinated or unsupported claims. Such inaccuracies risk spreading misinformation and diminishing customer trust. Second, without providing knowledge source attribution in CSA response, customers struggle to verify LLM-generated information. To address both challenges, we present an easily productionized solution that enables a ''citation experience'' to our customers. We build auto-evaluation metrics to holistically evaluate LLM's grounding and attribution capabilities, suggesting that citation generation paradigm substantially improves grounding performance by 13.83%. To deploy this capability at scale, we introduce Multi-UX-Inference system, which appends source citations to LLM outputs while preserving existing user experience features and supporting scalable inference. Large-scale online A/B tests show that grounded CSA responses improves customer engagement by 3% - 10%, depending on UX variations.

Fali Wang, Hui Liu, Zhenwei Dai et al.

Test-time scaling (TTS) enhances the performance of large language models (LLMs) by allocating additional compute resources during inference. However, existing research primarily investigates TTS in single-stage tasks; while many real-world problems are multi-stage complex tasks, composed of a sequence of heterogeneous subtasks with each subtask requires LLM of specific capability. Therefore, we study a novel problem: the test-time compute-optimal scaling in multi-stage complex tasks, aiming to select suitable models and allocate budgets per subtask to maximize overall performance. TTS in multi-stage tasks introduces two fundamental challenges: (i) The combinatorial search space of model and budget allocations, combined with the high cost of inference, makes brute-force search impractical. (ii) The optimal model and budget allocations across subtasks are interdependent, increasing the complexity of the compute-optimal search. To address this gap, we conduct extensive pilot experiments on four tasks across six datasets, deriving three empirical insights characterizing the behavior of LLMs in multi-stage complex tasks. Informed by these insights, we propose AgentTTS, an LLM-agent-based framework that autonomously searches for compute-optimal allocations through iterative feedback-driven interactions with the execution environment. Experimental results demonstrate that AgentTTS significantly outperforms traditional and other LLM-based baselines in search efficiency, and shows improved robustness to varying training set sizes and enhanced interpretability.

Jingying Zeng, Zhenwei Dai, Hui Liu et al.

Prompting LLMs offers an efficient way to guide output generation without explicit model training. In the e-commerce domain, prompting-based applications are widely used for tasks such as query understanding, recommender systems, and customer support. However, adapting LLMs to different tasks often requires extensive prompt engineering by domain experts, along with frequent updates to align with evolving business needs. Additionally, crafting fully unbiased natural language prompts remains a challenge for humans. To address these challenges, we propose a novel framework, Examples as the Prompt (EaP) which leverages labeled data to enhance prompts. Specifically, EaP automatically selects the most representative examples to maximize the few-shot capability of LLMs. It is efficient due to its unsupervised example selection and adaptive to potential data distribution shifts. We validate EaP on four real-world production use cases, demonstrating that it achieves comparable or even superior performance comparing to hand-crafted prompts designed by domain experts. Additionally, we introduce EaP_lite, which entirely replaces the natural language components of prompts with labeled examples. EaP_lite improves LLM inference speed by up to 70% without compromising performance. Latest online A/B test shows that using EaP and EaP_lite for data labeling can bring significant composite revenue gain by 0.06%.

Zhiwei Zhang, Hui Liu, Xiaomin Li et al.

Reward models trained on human preference data have demonstrated strong effectiveness in aligning Large Language Models (LLMs) with human intent under the framework of Reinforcement Learning from Human Feedback (RLHF). However, RLHF remains vulnerable to reward hacking, where the policy exploits imperfections in the reward function rather than genuinely learning the intended behavior. Although significant efforts have been made to mitigate reward hacking, they predominantly focus on and evaluate in-distribution scenarios, where the training and testing data for the reward model share the same distribution. In this paper, we empirically show that state-of-the-art methods struggle in more challenging out-of-distribution (OOD) settings. We further demonstrate that incorporating fine-grained multi-attribute scores helps address this challenge. However, the limited availability of high-quality data often leads to weak performance of multi-objective reward functions, which can negatively impact overall performance and become the bottleneck. To address this issue, we propose a unified reward modeling framework that jointly trains Bradley--Terry (BT) single-objective and multi-objective regression-based reward functions using a shared embedding space. We theoretically establish a connection between the BT loss and the regression objective and highlight their complementary benefits. Specifically, the regression task enhances the single-objective reward function's ability to mitigate reward hacking in challenging OOD settings, while BT-based training improves the scoring capability of the multi-objective reward function, enabling a 7B model to outperform a 70B baseline. Extensive experimental results demonstrate that our framework significantly improves both the robustness and the scoring performance of reward models.

Jingying Zeng, Jaewon Yang, Waleed Malik et al.

While generative AI excels in content generation, it does not always increase user engagement. This can be attributed to two main factors. First, generative AI generates content without incorporating explicit or implicit feedback about user interactions. Even if the generated content seems to be more informative or well-written, it does not necessarily lead to an increase in user activities, such as clicks. Second, there is a concern with the quality of the content generative AI produces, which often lacks the distinctiveness and authenticity that human-created content possesses. These two factors can lead to content that fails to meet specific needs and preferences of users, ultimately reducing its potential to be engaging. This paper presents a generic framework of how to improve user engagement with generative AI by leveraging user feedback. Our solutions employ rejection sampling, a technique used in reinforcement learning, to boost engagement metrics. We leveraged the framework in the context of email notification subject lines generation for an online social network, and achieved significant engagement metric lift including +1% Session and +0.4% Weekly Active Users. We believe our work offers a universal framework that enhances user engagement with generative AI, particularly when standard generative AI reaches its limits in terms of enhancing content to be more captivating. To the best of our knowledge, this represents an early milestone in the industry's successful use of generative AI to enhance user engagement.

Yingqian Cui, Zhenwei Dai, Pengfei He et al.

Large Language Models (LLMs) have achieved significant advances in reasoning tasks. A key approach is tree-based search with verifiers, which expand candidate reasoning paths and use reward models to guide pruning and selection. Although effective in improving accuracy, these methods are not optimal in terms of efficiency: they perform simple decomposition on the reasoning process, but ignore the planning-execution nature of tasks such as math reasoning or code generation. This results in inefficient exploration of reasoning process. To address this, we propose a dual-phase test-time scaling framework that explicitly separates reasoning into planning and execution, and performs search over the two phases individually. Specifically, we decompose reasoning trajectories and develop reward models for each phase, enabling the search to explore and prune plans and executions separately. We further introduce a dynamic budget allocation mechanism that adaptively redistributes sampling effort based on reward feedback, allowing early stopping on confident steps and reallocation of computation to more challenging parts of the reasoning process. Experiments on both mathematical reasoning and code generation benchmarks demonstrate that our approach consistently improves accuracy while reducing redundant computation.