1118 papers
This paper proposes the Adaptive Social Learning (ASL) framework, featuring the Adaptive Mode Policy Optimization (AMPO) algorithm, to enable language agents to dynamically switch between intuitive and deliberative reasoning modes based on context, thereby achieving superior task performance and token efficiency compared to existing methods like GPT-4o and GRPO.
This paper introduces "Talk to Your Slides," a high-efficiency slide editing agent that leverages language-driven structured data manipulation instead of visual perception to achieve faster, more accurate, and cost-effective text-centric and formatting modifications compared to Multimodal LLM-based GUI agents, supported by the newly proposed TSBench benchmark.
The paper introduces PC Agent-E, an efficient training framework that synthesizes diverse action decisions using Claude 3.7 Sonnet to augment a small set of 312 human trajectories, resulting in a model that significantly outperforms both human-only training and direct distillation on the WindowsAgentArena-V2 benchmark.
This paper introduces REFLEX, a metacognitive framework that empowers LLM-driven robotic agents to decompose skills, reflect on failures, and creatively synthesize novel solutions, thereby significantly enhancing their zero-shot and few-shot performance in complex multi-robot collaboration tasks.
This paper introduces BitBypass, a novel black-box jailbreak attack that exploits hyphen-separated bitstream camouflage to bypass the safety alignment of state-of-the-art large language models, demonstrating superior stealth and success rates compared to existing adversarial methods.
DiaBlo is a parameter-efficient fine-tuning method that updates only the diagonal blocks of model weight matrices, offering a simple, stable, and theoretically grounded alternative to LoRA that achieves competitive performance with comparable memory efficiency and training speed.
The paper introduces Go-Browse, a method that uses structured graph-based exploration to automatically collect a large-scale dataset of web agent trajectories, which, when used to fine-tune a 7B language model, achieves state-of-the-art performance on the WebArena benchmark.
This paper introduces HSSBench, a comprehensive multilingual benchmark featuring over 13,000 samples generated through a novel expert-agent collaboration pipeline, designed to evaluate and address the current limitations of Multimodal Large Language Models in handling the interdisciplinary and abstract reasoning tasks characteristic of the Humanities and Social Sciences.
MedXIAOHE is a state-of-the-art medical vision-language foundation model that leverages an entity-aware continual pretraining framework, reinforcement learning, and tool-augmented agentic training to achieve superior diagnostic reasoning, reliability, and performance across diverse medical benchmarks.
This paper introduces a novel surrogate model that simultaneously preserves both the empirical symbol frequency distributions (such as Zipf's law) and the long-range correlation structures of symbolic sequences like language and DNA by mapping fractional Gaussian noise onto the original histogram, thereby enabling the disentanglement of structural features and the testing of scaling law origins.
This paper introduces FeynTune, a suite of 20 specialized Large Language Models fine-tuned on High-Energy Physics arXiv abstracts that outperform both their base model and leading commercial LLMs in theoretical physics tasks, offering valuable insights for developing domain-specific AI in the field.
This study reveals that while ChatGPT temporarily boosts human creative performance, its use ultimately leads to a reversion to baseline creativity and a persistent homogenization of content, challenging the notion that generative AI enhances long-term human creativity.
This paper addresses the safety challenges of end-to-end conversational AI by surveying the problem landscape, proposing a value-sensitive design framework for release decisions, and providing a suite of tools to help researchers mitigate potential harms.
The paper introduces BERT, a novel bidirectional language representation model that leverages pre-training on unlabeled text to achieve state-of-the-art performance across a wide range of natural language processing tasks with minimal fine-tuning.
This paper introduces the Transformer, a novel neural network architecture that relies entirely on attention mechanisms while eliminating recurrence and convolutions, demonstrating superior translation quality, faster training times, and strong generalization to other tasks compared to existing state-of-the-art models.
This paper introduces two novel, computationally efficient model architectures for learning high-quality continuous word vector representations from massive datasets, which achieve state-of-the-art performance in measuring syntactic and semantic word similarities at a fraction of the previous computational cost.