Psychology research has shown that humans are poor at estimating their performance on tasks, tending towards underconfidence on easy tasks and overconfidence on difficult tasks. We examine three LLMs, Llama-3-70B-instruct, Claude-3-Sonnet, and GPT-4o, on a range of QA tasks of varying difficulty, and show that models exhibit subtle differences from human patterns of overconfidence: less sensitive to task difficulty, and when prompted to answer based on different personas -- e.g., expert vs layman, or different race, gender, and ages -- the models will respond with stereotypically biased confidence estimations even though their underlying answer accuracy remains the same. Based on these observations, we propose Answer-Free Confidence Estimation (AFCE) to improve confidence calibration and LLM interpretability in these settings. AFCE is a self-assessment method that employs two stages of prompting, first eliciting only confidence scores on questions, then asking separately for the answer. Experiments on the MMLU and GPQA datasets spanning subjects and difficulty show that this separation of tasks significantly reduces overconfidence and delivers more human-like sensitivity to task difficulty.

State Space Models (SSMs), like recent Mamba2, have achieved remarkable performance and received extensive attention. However, deploying Mamba2 on resource-constrained edge devices encounters many problems: severe outliers within the linear layer challenging the quantization, diverse and irregular element-wise tensor operations, and hardware-unfriendly nonlinear functions in the SSM block. To address these issues, this paper presents FastMamba, a dedicated accelerator on FPGA with hardware-algorithm co-design to promote the deployment efficiency of Mamba2. Specifically, we successfully achieve 8-bit quantization for linear layers through Hadamard transformation to eliminate outliers. Moreover, a hardware-friendly and fine-grained power-of-two quantization framework is presented for the SSM block and convolution layer, and a first-order linear approximation is developed to optimize the nonlinear functions. Based on the accurate algorithm quantization, we propose an accelerator that integrates parallel vector processing units, pipelined execution dataflow, and an efficient SSM Nonlinear Approximation Unit, which enhances computational efficiency and reduces hardware complexity. Finally, we evaluate FastMamba on Xilinx VC709 FPGA. For the input prefill task on Mamba2-130M, FastMamba achieves 68.80\times and 8.90\times speedup over Intel Xeon 4210R CPU and NVIDIA RTX 3090 GPU, respectively. In the output decode experiment with Mamba2-2.7B, FastMamba attains 6\times higher energy efficiency than RTX 3090 GPU.

Large language models (LLMs) integrated with autonomous agents hold significant potential for advancing scientific discovery through automated reasoning and task execution. However, applying LLM agents to drug discovery is still constrained by challenges such as large-scale multimodal data processing, limited task automation, and poor support for domain-specific tools. To overcome these limitations, we introduce DrugPilot, a LLM-based agent system with a parameterized reasoning architecture designed for end-to-end scientific workflows in drug discovery. DrugPilot enables multi-stage research processes by integrating structured tool use with a novel parameterized memory pool. The memory pool converts heterogeneous data from both public sources and user-defined inputs into standardized representations. This design supports efficient multi-turn dialogue, reduces information loss during data exchange, and enhances complex scientific decision-making. To support training and benchmarking, we construct a drug instruction dataset covering eight core drug discovery tasks. Under the Berkeley function-calling benchmark, DrugPilot significantly outperforms state-of-the-art agents such as ReAct and LoT, achieving task completion rates of 98.0%, 93.5%, and 64.0% for simple, multi-tool, and multi-turn scenarios, respectively. These results highlight DrugPilot's potential as a versatile agent framework for computational science domains requiring automated, interactive, and data-integrated reasoning.

Modern enterprise computing systems integrate numerous subsystems to resolve a common task by yielding emergent behavior. A widespread approach is using services implemented with Web technologies like REST or OpenAPI, which offer an interaction mechanism and service documentation standard, respectively. Each service represents a specific business functionality, allowing encapsulation and easier maintenance. Despite the reduced maintenance costs on an individual service level, increased integration complexity arises. Consequently, automated service composition approaches have arisen to mitigate this issue. Nevertheless, these approaches have not achieved high acceptance in practice due to their reliance on complex formal modeling. Within this Ph.D. thesis, we analyze the application of Large Language Models (LLMs) to automatically integrate the services based on a natural language input. The result is a reusable service composition, e.g., as program code. While not always generating entirely correct results, the result can still be helpful by providing integration engineers with a close approximation of a suitable solution, which requires little effort to become operational. Our research involves (i) introducing a software architecture for automated service composition using LLMs, (ii) analyzing Retrieval Augmented Generation (RAG) for service discovery, (iii) proposing a novel natural language query-based benchmark for service discovery, and (iv) extending the benchmark to complete service composition scenarios. We have presented our software architecture as Compositio Prompto, the analysis of RAG for service discovery, and submitted a proposal for the service discovery benchmark. Open topics are primarily the extension of the service discovery benchmark to service composition scenarios and the improvements of the service composition generation, e.g., using fine-tuning or LLM agents.

We present ZSE-Cap (Zero-Shot Ensemble for Captioning), our 4th place system in Event-Enriched Image Analysis (EVENTA) shared task on article-grounded image retrieval and captioning. Our zero-shot approach requires no finetuning on the competition's data. For retrieval, we ensemble similarity scores from CLIP, SigLIP, and DINOv2. For captioning, we leverage a carefully engineered prompt to guide the Gemma 3 model, enabling it to link high-level events from the article to the visual content in the image. Our system achieved a final score of 0.42002, securing a top-4 position on the private test set, demonstrating the effectiveness of combining foundation models through ensembling and prompting. Our code is available at https://github.com/ductai05/ZSE-Cap.

Large Language Models (LLMs) are widely used to generate plausible text on online platforms, without revealing the generation process. As users increasingly encounter such black-box outputs, detecting hallucinations has become a critical challenge. To address this challenge, we focus on developing a hallucination detection framework for black-box generators. Motivated by the observation that hallucinations, once introduced, tend to persist, we sample future contexts. The sampled future contexts provide valuable clues for hallucination detection and can be effectively integrated with various sampling-based methods. We extensively demonstrate performance improvements across multiple methods using our proposed sampling approach.

We introduce Kimi K2, a Mixture-of-Experts (MoE) large language model with 32 billion activated parameters and 1 trillion total parameters. We propose the MuonClip optimizer, which improves upon Muon with a novel QK-clip technique to address training instability while enjoying the advanced token efficiency of Muon. Based on MuonClip, K2 was pre-trained on 15.5 trillion tokens with zero loss spike. During post-training, K2 undergoes a multi-stage post-training process, highlighted by a large-scale agentic data synthesis pipeline and a joint reinforcement learning (RL) stage, where the model improves its capabilities through interactions with real and synthetic environments. Kimi K2 achieves state-of-the-art performance among open-source non-thinking models, with strengths in agentic capabilities. Notably, K2 obtains 66.1 on Tau2-Bench, 76.5 on ACEBench (En), 65.8 on SWE-Bench Verified, and 47.3 on SWE-Bench Multilingual -- surpassing most open and closed-sourced baselines in non-thinking settings. It also exhibits strong capabilities in coding, mathematics, and reasoning tasks, with a score of 53.7 on LiveCodeBench v6, 49.5 on AIME 2025, 75.1 on GPQA-Diamond, and 27.1 on OJBench, all without extended thinking. These results position Kimi K2 as one of the most capable open-source large language models to date, particularly in software engineering and agentic tasks. We release our base and post-trained model checkpoints to facilitate future research and applications of agentic intelligence.

Recent advances have enabled LLM-powered AI agents to autonomously execute complex tasks by combining language model reasoning with tools, memory, and web access. But can these systems be trusted to follow deployment policies in realistic environments, especially under attack? To investigate, we ran the largest public red-teaming competition to date, targeting 22 frontier AI agents across 44 realistic deployment scenarios. Participants submitted 1.8 million prompt-injection attacks, with over 60,000 successfully eliciting policy violations such as unauthorized data access, illicit financial actions, and regulatory noncompliance. We use these results to build the Agent Red Teaming (ART) benchmark - a curated set of high-impact attacks - and evaluate it across 19 state-of-the-art models. Nearly all agents exhibit policy violations for most behaviors within 10-100 queries, with high attack transferability across models and tasks. Importantly, we find limited correlation between agent robustness and model size, capability, or inference-time compute, suggesting that additional defenses are needed against adversarial misuse. Our findings highlight critical and persistent vulnerabilities in today's AI agents. By releasing the ART benchmark and accompanying evaluation framework, we aim to support more rigorous security assessment and drive progress toward safer agent deployment.

Aquaculture plays a vital role in global food security and coastal economies by providing sustainable protein sources. As the industry expands to meet rising demand, it faces growing challenges such as disease outbreaks, inefficient feeding practices, rising labor costs, logistical inefficiencies, and critical hatchery issues, including high mortality rates and poor water quality control. Although artificial intelligence has made significant progress, existing machine learning methods fall short of addressing the domain-specific complexities of aquaculture. To bridge this gap, we introduce AQUA, the first large language model (LLM) tailored for aquaculture, designed to support farmers, researchers, and industry practitioners. Central to this effort is AQUADAPT (Data Acquisition, Processing and Tuning), an Agentic Framework for generating and refining high-quality synthetic data using a combination of expert knowledge, largescale language models, and automated evaluation techniques. Our work lays the foundation for LLM-driven innovations in aquaculture research, advisory systems, and decision-making tools.

-- With rapid advances in code generation, reasoning, and problem-solving, Large Language Models (LLMs) are increasingly applied in robotics, most existing work focuses on high-level tasks such as task decomposition. A few studies have explored the use of LLMs in feedback controller design, however, these efforts are restricted to overly simplified systems, fixed-structure gain tuning, and lack real-world validation. To further investigate LLMs in automatic control, this work targets a key subfield: adaptive control. Inspired by the framework of model reference adaptive control (MRAC), we propose an LLMs-guided adaptive compensator framework that avoids designing controllers from scratch. Instead, the LLMs are prompted using the discrepancies between an unknown system and a reference system to design a compensator that aligns the response of the unknown system with that of the reference, thereby achieving adaptivity. Experiments evaluate five methods--LLM-guided adaptive compensator, LLM-guided adaptive controller, indirect adaptive control, learning-based adaptive control, and MRAC--on soft and humanoid robots, in both simulated and real-world environments. Results show that the LLMs-guided adaptive com-pensator outperforms traditional adaptive controllers and significantly reduces reasoning complexity compared to the LLMs-guided adaptive controller. The Lyapunov-based analysis and reasoning-path inspection demonstrate that the LLMs-guided adaptive compensator enables a more structured design process by transforming mathematical derivation into a reasoning task, while exhibiting strong generalizability, adaptability, and robustness. This study opens a new direction for applying LLMs in the field of automatic control, offering greater deployability and practicality compared to vision-language models.