Strategic decision-making involves interactive reasoning where agents adapt their choices in response to others, yet existing evaluations of large language models (LLMs) often emphasize Nash Equilibrium (NE) approximation, overlooking the mechanisms driving their strategic choices. To bridge this gap, we introduce an evaluation framework grounded in behavioral game theory, disentangling reasoning capability from contextual effects. Testing 22 state-of-the-art LLMs, we find that GPT-o3-mini, GPT-o1, and DeepSeek-R1 dominate most games yet also demonstrate that the model scale alone does not determine performance. In terms of prompting enhancement, Chain-of-Thought (CoT) prompting is not universally effective, as it increases strategic reasoning only for models at certain levels while providing limited gains elsewhere. Additionally, we investigate the impact of encoded demographic features on the models, observing that certain assignments impact the decision-making pattern. For instance, GPT-4o shows stronger strategic reasoning with female traits than males, while Gemma assigns higher reasoning levels to heterosexual identities compared to other sexual orientations, indicating inherent biases. These findings underscore the need for ethical standards and contextual alignment to balance improved reasoning with fairness.
We introduce SEKI, a novel large language model (LLM)-based neural architecture search (NAS) method. Inspired by the chain-of-thought (CoT) paradigm in modern LLMs, SEKI operates in two key stages: self-evolution and knowledge distillation. In the self-evolution stage, LLMs initially lack sufficient reference examples, so we implement an iterative refinement mechanism that enhances architectures based on performance feedback. Over time, this process accumulates a repository of high-performance architectures. In the knowledge distillation stage, LLMs analyze common patterns among these architectures to generate new, optimized designs. Combining these two stages, SEKI greatly leverages the capacity of LLMs on NAS and without requiring any domain-specific data. Experimental results show that SEKI achieves state-of-the-art (SOTA) performance across various datasets and search spaces while requiring only 0.05 GPU-days, outperforming existing methods in both efficiency and accuracy. Furthermore, SEKI demonstrates strong generalization capabilities, achieving SOTA-competitive results across multiple tasks.
Chinese spelling correction (CSC) is a crucial task that aims to correct character errors in Chinese text. While conventional CSC focuses on character substitution errors caused by mistyping, two other common types of character errors, missing and redundant characters, have received less attention. These errors are often excluded from CSC datasets during the annotation process or ignored during evaluation, even when they have been annotated. This issue limits the practicality of the CSC task. To address this issue, we introduce the task of General Chinese Character Error Correction (C2EC), which focuses on all three types of character errors. We construct a high-quality C2EC benchmark by combining and manually verifying data from CCTC and Lemon datasets. We extend the training-free prompt-free CSC method to C2EC by using Levenshtein distance for handling length changes and leveraging an additional prompt-based large language model (LLM) to improve performance. Experiments show that our method enables a 14B-parameter LLM to be on par with models nearly 50 times larger on both conventional CSC and C2EC tasks, without any fine-tuning.
Robust alignment guardrails for large language models are becoming increasingly important with their widespread application. In contrast to previous studies, we demonstrate that inference-time activation interventions can bypass safety alignments and effectively steer model generations towards harmful AI coordination for Llama 2. Our method applies fine-grained interventions at specific model subcomponents, particularly attention heads, using a simple binary choice probing strategy. These interventions then generalise to the open-ended generation setting effectively circumventing safety guardrails. We show that probing single attention heads is more effective than intervening on full layers and intervening on only four attention heads is comparable to supervised fine-tuning. We further show that only a few example completions are needed to compute effective steering directions, which is an advantage over classical fine-tuning. Our findings highlight the shortcomings of current alignment techniques. In addition, our results suggest that, at the attention head level, activations encode fine-grained linearly separable behaviors. Practically, the approach offers a straightforward methodology to steer large language model behaviour, which could be extended to diverse domains beyond safety requiring fine-grained control over the model output. The code and datasets for this study can be found on https://github.com/PaulDrm/targeted_intervention.
Mathematical reasoning is an increasingly important indicator of large language model (LLM) capabilities, yet we lack understanding of how LLMs process even simple mathematical tasks. To address this, we reverse engineer how three mid-sized LLMs compute addition. We first discover that numbers are represented in these LLMs as a generalized helix, which is strongly causally implicated for the tasks of addition and subtraction, and is also causally relevant for integer division, multiplication, and modular arithmetic. We then propose that LLMs compute addition by manipulating this generalized helix using the "Clock" algorithm: to solve $a+b$, the helices for $a$ and $b$ are manipulated to produce the $a+b$ answer helix which is then read out to model logits. We model influential MLP outputs, attention head outputs, and even individual neuron preactivations with these helices and verify our understanding with causal interventions. By demonstrating that LLMs represent numbers on a helix and manipulate this helix to perform addition, we present the first representation-level explanation of an LLM's mathematical capability.
The AgentSociety Challenge is the first competition in the Web Conference that aims to explore the potential of Large Language Model (LLM) agents in modeling user behavior and enhancing recommender systems on web platforms. The Challenge consists of two tracks: the User Modeling Track and the Recommendation Track. Participants are tasked to utilize a combined dataset from Yelp, Amazon, and Goodreads, along with an interactive environment simulator, to develop innovative LLM agents. The Challenge has attracted 295 teams across the globe and received over 1,400 submissions in total over the course of 37 official competition days. The participants have achieved 21.9% and 20.3% performance improvement for Track 1 and Track 2 in the Development Phase, and 9.1% and 15.9% in the Final Phase, representing a significant accomplishment. This paper discusses the detailed designs of the Challenge, analyzes the outcomes, and highlights the most successful LLM agent designs. To support further research and development, we have open-sourced the benchmark environment at https://tsinghua-fib-lab.github.io/AgentSocietyChallenge.
Computer vision is a critical component in a wide range of real-world applications, including plant monitoring in agriculture and handwriting classification in digital systems. However, developing high-performance computer vision models traditionally demands both machine learning (ML) expertise and domain-specific knowledge, making the process costly, labor-intensive, and inaccessible to many. Large language model (LLM) agents have emerged as a promising solution to automate this workflow, but most existing methods share a common limitation: they attempt to optimize entire pipelines in a single step before evaluation, making it difficult to attribute improvements to specific changes. This lack of granularity leads to unstable optimization and slower convergence, limiting their effectiveness. To address this, we introduce Iterative Refinement, a novel strategy for LLM-driven ML pipeline design inspired by how human ML experts iteratively refine models, focusing on one component at a time rather than making sweeping changes all at once. By systematically updating individual components based on real training feedback, Iterative Refinement improves stability, interpretability, and overall model performance. We implement this strategy in IMPROVE, an end-to-end LLM agent framework for automating and optimizing object classification pipelines. Through extensive evaluations across datasets of varying sizes and domains, including standard benchmarks and Kaggle competition datasets, we demonstrate that Iterative Refinement enables IMPROVE to consistently achieve better performance over existing zero-shot LLM-based approaches. These findings establish Iterative Refinement as an effective new strategy for LLM-driven ML automation and position IMPROVE as an accessible solution for building high-quality computer vision models without requiring ML expertise.
Building an agent that can mimic human behavior patterns to accomplish various open-world tasks is a long-term goal. To enable agents to effectively learn behavioral patterns across diverse tasks, a key challenge lies in modeling the intricate relationships among observations, actions, and language. To this end, we propose Optimus-2, a novel Minecraft agent that incorporates a Multimodal Large Language Model (MLLM) for high-level planning, alongside a Goal-Observation-Action Conditioned Policy (GOAP) for low-level control. GOAP contains (1) an Action-guided Behavior Encoder that models causal relationships between observations and actions at each timestep, then dynamically interacts with the historical observation-action sequence, consolidating it into fixed-length behavior tokens, and (2) an MLLM that aligns behavior tokens with open-ended language instructions to predict actions auto-regressively. Moreover, we introduce a high-quality Minecraft Goal-Observation-Action (MGOA)} dataset, which contains 25,000 videos across 8 atomic tasks, providing about 30M goal-observation-action pairs. The automated construction method, along with the MGOA dataset, can contribute to the community's efforts to train Minecraft agents. Extensive experimental results demonstrate that Optimus-2 exhibits superior performance across atomic tasks, long-horizon tasks, and open-ended instruction tasks in Minecraft.
This paper focuses on the piloting of CyberScholar, a Generative AI (GenAI) assistant tool that aims to provide feedback on writing K-12 contexts. The aim was to use GenAI to provide formative and summative feedback on students' texts in English Language Arts (ELA), Social Studies, and Modern World History. The trials discussed in this paper involved Grades 7, 8, 10, and 11 and were conducted in three schools in the Midwest and one in the Northwest of the United States. The tool used two main mechanisms: "prompt engineering" based on participant teachers' assessment rubric and "fine-tuning" a Large Language Model (LLM) from a customized corpus of teaching materials using Retrieval Augmented Generation. This paper focuses on CyberScholar's potential to enhance students' writing abilities and support teachers in diverse subject areas requiring written assignments.
Recently, the number of off-the-shelf Large Language Models (LLMs) has exploded with many open-source options. This creates a diverse landscape regarding both serving options (e.g., inference on local hardware vs remote LLM APIs) and model heterogeneous expertise. However, it is hard for the user to efficiently optimize considering operational cost (pricing structures, expensive LLMs-as-a-service for large querying volumes), efficiency, or even per-case specific measures such as response accuracy, bias, or toxicity. Also, existing LLM routing solutions focus mainly on cost reduction, with response accuracy optimizations relying on non-generalizable supervised training, and ensemble approaches necessitating output computation for every considered LLM candidate. In this work, we tackle the challenge of selecting the optimal LLM from a model pool for specific queries with customizable objectives. We propose PickLLM, a lightweight framework that relies on Reinforcement Learning (RL) to route on-the-fly queries to available models. We introduce a weighted reward function that considers per-query cost, inference latency, and model response accuracy by a customizable scoring function. Regarding the learning algorithms, we explore two alternatives: PickLLM router acting as a learning automaton that utilizes gradient ascent to select a specific LLM, or utilizing stateless Q-learning to explore the set of LLMs and perform selection with a $\epsilon$-greedy approach. The algorithm converges to a single LLM for the remaining session queries. To evaluate, we utilize a pool of four LLMs and benchmark prompt-response datasets with different contexts. A separate scoring function is assessing response accuracy during the experiment. We demonstrate the speed of convergence for different learning rates and improvement in hard metrics such as cost per querying session and overall response latency.
Recent developments in large language models (LLMs) have led to their widespread usage for various tasks. The prevalence of LLMs in society implores the assurance on the reliability of their performance. In particular, risk-sensitive applications demand meticulous attention to unexpectedly poor outcomes, i.e., tail events, for instance, toxic answers, humiliating language, and offensive outputs. Due to the costly nature of acquiring human annotations, general-purpose scoring models have been created to automate the process of quantifying these tail events. This phenomenon introduces potential human-machine misalignment between the respective scoring mechanisms. In this work, we present a lightweight calibration framework for blackbox models that ensures the alignment of humans and machines with provable guarantees. Our framework provides a rigorous approach to controlling any distortion risk measure that is characterized by a weighted average of quantiles of the loss incurred by the LLM with high confidence. The theoretical foundation of our method relies on the connection between conformal risk control and a traditional family of statistics, i.e., L-statistics. To demonstrate the utility of our framework, we conduct comprehensive experiments that address the issue of human-machine misalignment.
Effective instruction fine-tuning on diverse image-text datasets is crucial for developing a versatile Multimodal Large Language Model (MLLM), where dataset composition dictates the model's adaptability across multimodal tasks. However, complex datasets often contain inherent conflicts -- stemming from modality-specific optimization objectives -- and latent commonalities that enable cross-task transfer, which most existing approaches handle separately. To bridge this gap, we introduce AsymLoRA, a parameter-efficient tuning framework that unifies knowledge modularization and cross-modal coordination via asymmetric LoRA: task-specific low-rank projections (matrix B) that preserve distinct adaptation pathways for conflicting objectives, and a shared projection (matrix A) that consolidates cross-modal commonalities. Extensive evaluations demonstrate that AsymLoRA consistently surpasses both vanilla LoRA, which captures only commonalities, and LoRA-MoE, which focuses solely on conflicts, achieving superior model performance and system efficiency across diverse benchmarks.\href{Code}{https://github.com/Clin0212/HydraLoRA/blob/main/MLLM-HydraLoRA/README.md}.
Recent advances in Multi-Modal Large Language Models (M-LLMs) show promising results in video reasoning. Popular Multi-Modal Large Language Model (M-LLM) frameworks usually apply naive uniform sampling to reduce the number of video frames that are fed into an M-LLM, particularly for long context videos. However, it could lose crucial context in certain periods of a video, so that the downstream M-LLM may not have sufficient visual information to answer a question. To attack this pain point, we propose a light-weight M-LLM -based frame selection method that adaptively select frames that are more relevant to users' queries. In order to train the proposed frame selector, we introduce two supervision signals (i) Spatial signal, where single frame importance score by prompting a M-LLM; (ii) Temporal signal, in which multiple frames selection by prompting Large Language Model (LLM) using the captions of all frame candidates. The selected frames are then digested by a frozen downstream video M-LLM for visual reasoning and question answering. Empirical results show that the proposed M-LLM video frame selector improves the performances various downstream video Large Language Model (video-LLM) across medium (ActivityNet, NExT-QA) and long (EgoSchema, LongVideoBench) context video question answering benchmarks.
Drug discovery is a critical task in biomedical natural language processing (NLP), yet explainable drug discovery remains underexplored. Meanwhile, large language models (LLMs) have shown remarkable abilities in natural language understanding and generation. Leveraging LLMs for explainable drug discovery has the potential to improve downstream tasks and real-world applications. In this study, we utilize open-source drug knowledge graphs, clinical trial data, and PubMed publications to construct a comprehensive dataset for the explainable drug discovery task, named \textbf{expRxRec}. Furthermore, we introduce \textbf{KEDRec-LM}, an instruction-tuned LLM which distills knowledge from rich medical knowledge corpus for drug recommendation and rationale generation. To encourage further research in this area, we will publicly release\footnote{A copy is attached with this submission} both the dataset and KEDRec-LM.
Pok\'emon Red, a classic Game Boy JRPG, presents significant challenges as a testbed for agents, including multi-tasking, long horizons of tens of thousands of steps, hard exploration, and a vast array of potential policies. We introduce a simplistic environment and a Deep Reinforcement Learning (DRL) training methodology, demonstrating a baseline agent that completes an initial segment of the game up to completing Cerulean City. Our experiments include various ablations that reveal vulnerabilities in reward shaping, where agents exploit specific reward signals. We also discuss limitations and argue that games like Pok\'emon hold strong potential for future research on Large Language Model agents, hierarchical training algorithms, and advanced exploration methods. Source Code: https://github.com/MarcoMeter/neroRL/tree/poke_red
Machine learning systems trained on electronic health records (EHRs) increasingly guide treatment decisions, but their reliability depends on the critical assumption that patients follow the prescribed treatments recorded in EHRs. Using EHR data from 3,623 hypertension patients, we investigate how treatment non-adherence introduces implicit bias that can fundamentally distort both causal inference and predictive modeling. By extracting patient adherence information from clinical notes using a large language model (LLM), we identify 786 patients (21.7%) with medication non-adherence. We further uncover key demographic and clinical factors associated with non-adherence, as well as patient-reported reasons including side effects and difficulties obtaining refills. Our findings demonstrate that this implicit bias can not only reverse estimated treatment effects, but also degrade model performance by up to 5% while disproportionately affecting vulnerable populations by exacerbating disparities in decision outcomes and model error rates. This highlights the importance of accounting for treatment non-adherence in developing responsible and equitable clinical machine learning systems.
The multivariate, asynchronous nature of real-world clinical data, such as that generated in Intensive Care Units (ICUs), challenges traditional AI-based decision-support systems. These often assume data regularity and feature independence and frequently rely on limited data scopes and manual feature engineering. The potential of generative AI technologies has not yet been fully exploited to analyze clinical data. We introduce ICU-BERT, a transformer-based model pre-trained on the MIMIC-IV database using a multi-task scheme to learn robust representations of complex ICU data with minimal preprocessing. ICU-BERT employs a multi-token input strategy, incorporating dense embeddings from a biomedical Large Language Model to learn a generalizable representation of complex and multivariate ICU data. With an initial evaluation of five tasks and four additional ICU datasets, ICU-BERT results indicate that ICU-BERT either compares to or surpasses current performance benchmarks by leveraging fine-tuning. By integrating structured and unstructured data, ICU-BERT advances the use of foundational models in medical informatics, offering an adaptable solution for clinical decision support across diverse applications.
Stance detection on social media aims to identify attitudes expressed in tweets towards specific targets. Current studies prioritize Large Language Models (LLMs) over Small Language Models (SLMs) due to the overwhelming performance improving provided by LLMs. However, heavily relying on LLMs for stance detection, regardless of the cost, is impractical for real-world social media monitoring systems that require vast data analysis. To this end, we propose \textbf{\underline{Co}}llaborative Stance Detection via Small-Large Language Model Consistency \textbf{\underline{Ver}}ification (\textbf{CoVer}) framework, which enhances LLM utilization via context-shared batch reasoning and logical verification between LLM and SLM. Specifically, instead of processing each text individually, CoVer processes texts batch-by-batch, obtaining stance predictions and corresponding explanations via LLM reasoning in a shared context. Then, to exclude the bias caused by context noises, CoVer introduces the SLM for logical consistency verification. Finally, texts that repeatedly exhibit low logical consistency are classified using consistency-weighted aggregation of prior LLM stance predictions. Our experiments show that CoVer outperforms state-of-the-art methods across multiple benchmarks in the zero-shot setting, achieving 0.54 LLM queries per tweet while significantly enhancing performance. Our CoVer offers a more practical solution for LLM deploying for social media stance detection.
Large language models (LLMs) have shown promise in various natural language processing tasks, including their application to proteomics data to classify protein fragments. In this study, we curated a limited mass spectrometry dataset with 1000s of protein fragments, consisting of proteins that appear to be attached to the endoplasmic reticulum in cardiac cells, of which a fraction was cloned and characterized for their impact on SERCA, an ER calcium pump. With this limited dataset, we sought to determine whether LLMs could correctly predict whether a new protein fragment could bind SERCA, based only on its sequence and a few biophysical characteristics, such as hydrophobicity, determined from that sequence. To do so, we generated random sequences based on cloned fragments, embedded the fragments into a retrieval augmented generation (RAG) database to group them by similarity, then fine-tuned large language model (LLM) prompts to predict whether a novel sequence could bind SERCA. We benchmarked this approach using multiple open-source LLMs, namely the Meta/llama series, and embedding functions commonly available on the Huggingface repository. We then assessed the generalizability of this approach in classifying novel protein fragments from mass spectrometry that were not initially cloned for functional characterization. By further tuning the prompt to account for motifs, such as ER retention sequences, we improved the classification accuracy by and identified several proteins predicted to localize to the endoplasmic reticulum and bind SERCA, including Ribosomal Protein L2 and selenoprotein S. Although our results were based on proteomics data from cardiac cells, our approach demonstrates the potential of LLMs in identifying novel protein interactions and functions with very limited proteomic data.
Transformers consist of diverse building blocks, such as embedding layers, normalization layers, self-attention mechanisms, and point-wise feedforward networks. Thus, understanding the differences and interactions among these blocks is important. In this paper, we uncover a clear Sharpness Disparity across these blocks, which emerges early in training and intriguingly persists throughout the training process. Motivated by this finding, we propose Blockwise Learning Rate (LR), a strategy that tailors the LR to each block's sharpness, accelerating large language model (LLM) pre-training. By integrating Blockwise LR into AdamW, we consistently achieve lower terminal loss and nearly $2\times$ speedup compared to vanilla AdamW. We demonstrate this acceleration across GPT-2 and LLaMA, with model sizes ranging from 0.12B to 2B and datasets of OpenWebText, MiniPile, and C4. Finally, we incorporate Blockwise LR into Adam-mini (Zhang et al., 2024), a recently proposed memory-efficient variant of Adam, achieving a combined $2\times$ speedup and $2\times$ memory saving. These results underscore the potential of exploiting the sharpness disparity to improve LLM training.