Providing high-quality feedback is crucial for student success but is constrained by time, cost, and limited data availability. We introduce Synthetic Educational Feedback Loops (SEFL), a novel framework designed to deliver immediate, on-demand feedback at scale without relying on extensive, real-world student data. In SEFL, two large language models (LLMs) operate in teacher--student roles to simulate assignment completion and formative feedback, generating abundant synthetic pairs of student work and corresponding critiques. We then fine-tune smaller, more computationally efficient LLMs on these synthetic pairs, enabling them to replicate key features of high-quality, goal-oriented feedback. Unlike personalized tutoring approaches that offer multi-turn, individualized instruction, SEFL specifically focuses on replicating the teacher-->student feedback loop for diverse assignments. Through both LLM-as-a-judge and human evaluations, we demonstrate that SEFL-tuned models outperform their non-tuned counterparts in feedback quality, clarity, and timeliness. These findings reveal SEFL's potential to transform feedback processes for higher education and beyond, offering an ethical and scalable alternative to conventional manual feedback cycles.
Large Language Model (LLM)-based agents have demonstrated remarkable success in solving complex tasks across a wide range of general-purpose applications. However, their performance often degrades in context-specific scenarios, such as specialized industries or research domains, where the absence of domain-relevant knowledge leads to imprecise or suboptimal outcomes. To address this challenge, our work introduces a systematic approach to enhance the contextual adaptability of LLM-based agents by optimizing their underlying prompts-critical components that govern agent behavior, roles, and interactions. Manually crafting optimized prompts for context-specific tasks is labor-intensive, error-prone, and lacks scalability. In this work, we introduce an Extractor-Generator framework designed to automate the optimization of contextual LLM-based agents. Our method operates through two key stages: (i) feature extraction from a dataset of gold-standard input-output examples, and (ii) prompt generation via a high-level optimization strategy that iteratively identifies underperforming cases and applies self-improvement techniques. This framework substantially improves prompt adaptability by enabling more precise generalization across diverse inputs, particularly in context-specific tasks where maintaining semantic consistency and minimizing error propagation are critical for reliable performance. Although developed with single-stage workflows in mind, the approach naturally extends to multi-stage workflows, offering broad applicability across various agent-based systems. Empirical evaluations demonstrate that our framework significantly enhances the performance of prompt-optimized agents, providing a structured and efficient approach to contextual LLM-based agents.
Current multimodal information retrieval studies mainly focus on single-image inputs, which limits real-world applications involving multiple images and text-image interleaved content. In this work, we introduce the text-image interleaved retrieval (TIIR) task, where the query and document are interleaved text-image sequences, and the model is required to understand the semantics from the interleaved context for effective retrieval. We construct a TIIR benchmark based on naturally interleaved wikiHow tutorials, where a specific pipeline is designed to generate interleaved queries. To explore the task, we adapt several off-the-shelf retrievers and build a dense baseline by interleaved multimodal large language model (MLLM). We then propose a novel Matryoshka Multimodal Embedder (MME), which compresses the number of visual tokens at different granularity, to address the challenge of excessive visual tokens in MLLM-based TIIR models. Experiments demonstrate that simple adaption of existing models does not consistently yield effective results. Our MME achieves significant improvements over the baseline by substantially fewer visual tokens. We provide extensive analysis and will release the dataset and code to facilitate future research.
Explainable recommendation has demonstrated significant advantages in informing users about the logic behind recommendations, thereby increasing system transparency, effectiveness, and trustworthiness. To provide personalized and interpretable explanations, existing works often combine the generation capabilities of large language models (LLMs) with collaborative filtering (CF) information. CF information extracted from the user-item interaction graph captures the user behaviors and preferences, which is crucial for providing informative explanations. However, due to the complexity of graph structure, effectively extracting the CF information from graphs still remains a challenge. Moreover, existing methods often struggle with the integration of extracted CF information with LLMs due to its implicit representation and the modality gap between graph structures and natural language explanations. To address these challenges, we propose G-Refer, a framework using graph retrieval-augmented large language models (LLMs) for explainable recommendation. Specifically, we first employ a hybrid graph retrieval mechanism to retrieve explicit CF signals from both structural and semantic perspectives. The retrieved CF information is explicitly formulated as human-understandable text by the proposed graph translation and accounts for the explanations generated by LLMs. To bridge the modality gap, we introduce knowledge pruning and retrieval-augmented fine-tuning to enhance the ability of LLMs to process and utilize the retrieved CF information to generate explanations. Extensive experiments show that G-Refer achieves superior performance compared with existing methods in both explainability and stability. Codes and data are available at https://github.com/Yuhan1i/G-Refer.
Understanding how individuals perceive and recall information in their natural environments is critical to understanding potential failures in perception (e.g., sensory loss) and memory (e.g., dementia). Event segmentation, the process of identifying distinct events within dynamic environments, is central to how we perceive, encode, and recall experiences. This cognitive process not only influences moment-to-moment comprehension but also shapes event specific memory. Despite the importance of event segmentation and event memory, current research methodologies rely heavily on human judgements for assessing segmentation patterns and recall ability, which are subjective and time-consuming. A few approaches have been introduced to automate event segmentation and recall scoring, but validity with human responses and ease of implementation require further advancements. To address these concerns, we leverage Large Language Models (LLMs) to automate event segmentation and assess recall, employing chat completion and text-embedding models, respectively. We validated these models against human annotations and determined that LLMs can accurately identify event boundaries, and that human event segmentation is more consistent with LLMs than among humans themselves. Using this framework, we advanced an automated approach for recall assessments which revealed semantic similarity between segmented narrative events and participant recall can estimate recall performance. Our findings demonstrate that LLMs can effectively simulate human segmentation patterns and provide recall evaluations that are a scalable alternative to manual scoring. This research opens novel avenues for studying the intersection between perception, memory, and cognitive impairment using methodologies driven by artificial intelligence.
We explore how private synthetic text can be generated by suitably prompting a large language model (LLM). This addresses a challenge for organizations like hospitals, which hold sensitive text data like patient medical records, and wish to share it in order to train machine learning models for medical tasks, while preserving patient privacy. Methods that rely on training or finetuning a model may be out of reach, either due to API limits of third-party LLMs, or due to ethical and legal prohibitions on sharing the private data with the LLM itself. We propose Differentially Private Keyphrase Prompt Seeding (DP-KPS), a method that generates a private synthetic text corpus from a sensitive input corpus, by accessing an LLM only through privatized prompts. It is based on seeding the prompts with private samples from a distribution over phrase embeddings, thus capturing the input corpus while achieving requisite output diversity and maintaining differential privacy. We evaluate DP-KPS on downstream ML text classification tasks, and show that the corpora it generates preserve much of the predictive power of the original ones. Our findings offer hope that institutions can reap ML insights by privately sharing data with simple prompts and little compute.
How should one judge whether a given large language model (LLM) can reliably perform economic reasoning? Most existing LLM benchmarks focus on specific applications and fail to present the model with a rich variety of economic tasks. A notable exception is Raman et al. [2024], who offer an approach for comprehensively benchmarking strategic decision-making; however, this approach fails to address the non-strategic settings prevalent in microeconomics, such as supply-and-demand analysis. We address this gap by taxonomizing microeconomic reasoning into $58$ distinct elements, focusing on the logic of supply and demand, each grounded in up to $10$ distinct domains, $5$ perspectives, and $3$ types. The generation of benchmark data across this combinatorial space is powered by a novel LLM-assisted data generation protocol that we dub auto-STEER, which generates a set of questions by adapting handwritten templates to target new domains and perspectives. Because it offers an automated way of generating fresh questions, auto-STEER mitigates the risk that LLMs will be trained to over-fit evaluation benchmarks; we thus hope that it will serve as a useful tool both for evaluating and fine-tuning models for years to come. We demonstrate the usefulness of our benchmark via a case study on $27$ LLMs, ranging from small open-source models to the current state of the art. We examined each model's ability to solve microeconomic problems across our whole taxonomy and present the results across a range of prompting strategies and scoring metrics.
When encountering increasingly frequent performance improvements or cost reductions from a new large language model (LLM), developers of applications leveraging LLMs must decide whether to take advantage of these improvements or stay with older tried-and-tested models. Low perceived switching frictions can lead to choices that do not consider more subtle behavior changes that the transition may induce. Our experiments use a popular game-theoretic behavioral economics model of trust to show stark differences in the trusting behavior of OpenAI's and DeepSeek's models. We highlight a collapse in the economic trust behavior of the o1-mini and o3-mini models as they reconcile profit-maximizing and risk-seeking with future returns from trust, and contrast it with DeepSeek's more sophisticated and profitable trusting behavior that stems from an ability to incorporate deeper concepts like forward planning and theory-of-mind. As LLMs form the basis for high-stakes commercial systems, our results highlight the perils of relying on LLM performance benchmarks that are too narrowly defined and suggest that careful analysis of their hidden fault lines should be part of any organization's AI strategy.
This paper presents DataSciBench, a comprehensive benchmark for evaluating Large Language Model (LLM) capabilities in data science. Recent related benchmarks have primarily focused on single tasks, easily obtainable ground truth, and straightforward evaluation metrics, which limits the scope of tasks that can be evaluated. In contrast, DataSciBench is constructed based on a more comprehensive and curated collection of natural and challenging prompts for uncertain ground truth and evaluation metrics. We develop a semi-automated pipeline for generating ground truth (GT) and validating evaluation metrics. This pipeline utilizes and implements an LLM-based self-consistency and human verification strategy to produce accurate GT by leveraging collected prompts, predefined task types, and aggregate functions (metrics). Furthermore, we propose an innovative Task - Function - Code (TFC) framework to assess each code execution outcome based on precisely defined metrics and programmatic rules. Our experimental framework involves testing 6 API-based models, 8 open-source general models, and 9 open-source code generation models using the diverse set of prompts we have gathered. This approach aims to provide a more comprehensive and rigorous evaluation of LLMs in data science, revealing their strengths and weaknesses. Experimental results demonstrate that API-based models outperform open-sourced models on all metrics and Deepseek-Coder-33B-Instruct achieves the highest score among open-sourced models. We release all code and data at https://github.com/THUDM/DataSciBench.
LLM-based user agents, which simulate user interaction behavior, are emerging as a promising approach to enhancing recommender systems. In real-world scenarios, users' interactions often exhibit cross-domain characteristics and are influenced by others. However, the memory design in current methods causes user agents to introduce significant irrelevant information during decision-making in cross-domain scenarios and makes them unable to recognize the influence of other users' interactions, such as popularity factors. To tackle this issue, we propose a dual-layer memory architecture combined with a two-step fusion mechanism. This design avoids irrelevant information during decision-making while ensuring effective integration of cross-domain preferences. We also introduce the concepts of interest groups and group-shared memory to better capture the influence of popularity factors on users with similar interests. Comprehensive experiments validate the effectiveness of AgentCF++. Our code is available at https://github.com/jhliu0807/AgentCF-plus.
Hallucinations in large language model (LLM) outputs severely limit their reliability in knowledge-intensive tasks such as question answering. To address this challenge, we introduce REFIND (Retrieval-augmented Factuality hallucINation Detection), a novel framework that detects hallucinated spans within LLM outputs by directly leveraging retrieved documents. As part of the REFIND, we propose the Context Sensitivity Ratio (CSR), a novel metric that quantifies the sensitivity of LLM outputs to retrieved evidence. This innovative approach enables REFIND to efficiently and accurately detect hallucinations, setting it apart from existing methods. In the evaluation, REFIND demonstrated robustness across nine languages, including low-resource settings, and significantly outperformed baseline models, achieving superior IoU scores in identifying hallucinated spans. This work highlights the effectiveness of quantifying context sensitivity for hallucination detection, thereby paving the way for more reliable and trustworthy LLM applications across diverse languages.
Ontology, and more broadly, Knowledge Graph Matching is a challenging task in which expressiveness has not been fully addressed. Despite the increasing use of embeddings and language models for this task, approaches for generating expressive correspondences still do not take full advantage of these models, in particular, large language models (LLMs). This paper proposes to integrate LLMs into an approach for generating expressive correspondences based on alignment need and ABox-based relation discovery. The generation of correspondences is performed by matching similar surroundings of instance sub-graphs. The integration of LLMs results in different architectural modifications, including label similarity, sub-graph matching, and entity matching. The performance word embeddings, sentence embeddings, and LLM-based embeddings, was compared. The results demonstrate that integrating LLMs surpasses all other models, enhancing the baseline version of the approach with a 45\% increase in F-measure.
Data augmentation is necessary for graph representation learning due to the scarcity and noise present in graph data. Most of the existing augmentation methods overlook the context information inherited from the dataset as they rely solely on the graph structure for augmentation. Despite the success of some large language model-based (LLM) graph learning methods, they are mostly white-box which require access to the weights or latent features from the open-access LLMs, making them difficult to be democratized for everyone as existing LLMs are mostly closed-source for commercial considerations. To overcome these limitations, we propose a black-box context-driven graph data augmentation approach, with the guidance of LLMs -- DemoGraph. Leveraging the text prompt as context-related information, we task the LLM with generating knowledge graphs (KGs), which allow us to capture the structural interactions from the text outputs. We then design a dynamic merging schema to stochastically integrate the LLM-generated KGs into the original graph during training. To control the sparsity of the augmented graph, we further devise a granularity-aware prompting strategy and an instruction fine-tuning module, which seamlessly generates text prompts according to different granularity levels of the dataset. Extensive experiments on various graph learning tasks validate the effectiveness of our method over existing graph data augmentation methods. Notably, our approach excels in scenarios involving electronic health records (EHRs), which validates its maximal utilization of contextual knowledge, leading to enhanced predictive performance and interpretability.
We show that Large Language Model from Power Law Decoder Representations (PLDR-LLM) is a foundational model whose deductive outputs are invariant tensors up to a small perturbation. PLDR-LLM learns a singularity condition for the deductive outputs that enable the once-inferred energy-curvature tensor $\mathbf{G}_{LM}$ to replace the deep neural network of power law graph attention (PLGA) generating the deductive outputs at inference. We demonstrate that a cache for $\mathbf{G}_{LM}$ (G-cache) and KV-cache can be implemented in a straightforward manner to improve the inference time. The invariance and generalizable nature of deductive outputs is at a very high fidelity where deductive outputs have same RMSE and determinant values up to 15 decimal places after caching, and zero-shot benchmark scores remain unchanged. Ablation studies show that learned deductive outputs have distinct loss and accuracy characteristics from models pretrained with transferred, randomly initialized or identity tensors as a constant tensor operator and an LLM with scaled-dot product attention (SDPA) is a special case of PLDR-LLM where $\mathbf{G}_{LM}$ is predefined as identity. The observed invariance characteristic introduces a novel asymmetry between training and inference phases with caching. We outline observed common characteristics of the deductive outputs for the learned singularity condition. We provide an implementation of a training and inference framework for PLDR-LLM with KV-cache and G-cache.
Large language model (LLM) systems suffer from the models' unstable ability to generate valid and factual content, resulting in hallucination generation. Current hallucination detection methods heavily rely on out-of-model information sources, such as RAG to assist the detection, thus bringing heavy additional latency. Recently, internal states of LLMs' inference have been widely used in numerous research works, such as prompt injection detection, etc. Considering the interpretability of LLM internal states and the fact that they do not require external information sources, we introduce such states into LLM hallucination detection. In this paper, we systematically analyze different internal states' revealing features during inference forward and comprehensively evaluate their ability in hallucination detection. Specifically, we cut the forward process of a large language model into three stages: understanding, query, generation, and extracting the internal state from these stages. By analyzing these states, we provide a deep understanding of why the hallucinated content is generated and what happened in the internal state of the models. Then, we introduce these internal states into hallucination detection and conduct comprehensive experiments to discuss the advantages and limitations.
It is popular lately to train large language models to be used as chat assistants, but in the conversation between the user and the chat assistant, there are prompts, require multi-turns between the chat assistant and the user. However, there are a number of issues with the multi-turns conversation: The response of the chat assistant is prone to errors and cannot help users achieve their goals; It is difficult for chat assistant to generate responses with different processes based on actual needs for the same command or request; Chat assistant require the use of tools, but the current approach is not elegant and efficient, and the number of tool calls that can be supported is limited. The main reason for these issues is that large language models do not have the thinking ability as a human, lack the reasoning ability and planning ability, and lack the ability to execute plans. To solve these issues, we propose a thinking method based on a built-in chain of thought: In the multi-turns conversation, for each user prompt, the large language model thinks based on elements such as chat history, thinking context, action calls, memory and knowledge, makes detailed reasoning and planning, and actions according to the plan. We also explored how the large language model enhances thinking ability through this thinking method: Collect training datasets according to the thinking method and fine tune the large language model through supervised learning; Train a consistency reward model and use it as a reward function to fine tune the large language model using reinforcement learning, and the reinforced large language model outputs according to this way of thinking. Our experimental results show that the reasoning ability and planning ability of the large language model are enhanced, and the issues in the multi-turns conversation are solved.
Guiding the policy of multi-agent reinforcement learning to align with human common sense is a difficult problem, largely due to the complexity of modeling common sense as a reward, especially in complex and long-horizon multi-agent tasks. Recent works have shown the effectiveness of reward shaping, such as potential-based rewards, to enhance policy alignment. The existing works, however, primarily rely on experts to design rule-based rewards, which are often labor-intensive and lack a high-level semantic understanding of common sense. To solve this problem, we propose a hierarchical vision-based reward shaping method. At the bottom layer, a visual-language model (VLM) serves as a generic potential function, guiding the policy to align with human common sense through its intrinsic semantic understanding. To help the policy adapts to uncertainty and changes in long-horizon tasks, the top layer features an adaptive skill selection module based on a visual large language model (vLLM). The module uses instructions, video replays, and training records to dynamically select suitable potential function from a pre-designed pool. Besides, our method is theoretically proven to preserve the optimal policy. Extensive experiments conducted in the Google Research Football environment demonstrate that our method not only achieves a higher win rate but also effectively aligns the policy with human common sense.
StarCraft II is a complex and dynamic real-time strategy (RTS) game environment, which is very suitable for artificial intelligence and reinforcement learning research. To address the problem of Large Language Model(LLM) learning in complex environments through self-reflection, we propose a Reflection of Episodes(ROE) framework based on expert experience and self-experience. This framework first obtains key information in the game through a keyframe selection method, then makes decisions based on expert experience and self-experience. After a game is completed, it reflects on the previous experience to obtain new self-experience. Finally, in the experiment, our method beat the robot under the Very Hard difficulty in TextStarCraft II. We analyze the data of the LLM in the process of the game in detail, verified its effectiveness.
Medical question answering requires extensive access to specialized conceptual knowledge. The current paradigm, Retrieval-Augmented Generation (RAG), acquires expertise medical knowledge through large-scale corpus retrieval and uses this knowledge to guide a general-purpose large language model (LLM) for generating answers. However, existing retrieval approaches often overlook the importance of factual knowledge, which limits the relevance of retrieved conceptual knowledge and restricts its applicability in real-world scenarios, such as clinical decision-making based on Electronic Health Records (EHRs). This paper introduces RGAR, a recurrence generation-augmented retrieval framework that retrieves both relevant factual and conceptual knowledge from dual sources (i.e., EHRs and the corpus), allowing them to interact and refine each another. Through extensive evaluation across three factual-aware medical question answering benchmarks, RGAR establishes a new state-of-the-art performance among medical RAG systems. Notably, the Llama-3.1-8B-Instruct model with RGAR surpasses the considerably larger, RAG-enhanced GPT-3.5. Our findings demonstrate the benefit of extracting factual knowledge for retrieval, which consistently yields improved generation quality.
Direct Preference Optimization (DPO) demonstrates the advantage of aligning a large language model with human preference using only an offline dataset. However, DPO has the limitation that the KL penalty, which prevents excessive deviation from the reference model, is static throughout the training process. Several methods claim to change this static KL penalty of DPO into a dynamic one, but no approach can adaptively assign different KL penalties for each preference pair. In this paper, we propose $\varepsilon$-Direct Preference Optimization ($\varepsilon$-DPO), which allows adaptive control of the KL penalty strength $\beta$ for each preference pair. Specifically, $\varepsilon$-DPO adaptively controls $\beta$ for each preference pair based on the monotonicity of logits as a preference model under the perturbation of $\beta$ during training. This is equivalent to adjusting the KL penalty by checking whether the change in training-time temperature can lead to better preference confidence as preference models by simply reusing the logit of the current policy and the reference policy. Experimental results show that the simple criterion of $\varepsilon$-DPO for KL penalty relaxation significantly improves DPO compared to most existing direct alignment algorithms on general chatbot benchmarks and reveal that this KL penalty control criterion can reflect confusion as a preference model and provide an efficient KL trade-off, highlighting the significance of instance-level adaptive KL penalty control in DPO.