Sloth: scaling laws for LLM skills to predict multi-benchmark performance across families

The recent introduction of DESIGNER, a design-logic-guided reasoning data synthesis pipeline, aims to enhance the capabilities of large language models (LLMs) in tackling complex, multidisciplinary questions. By leveraging extensive raw documents, DESIGNER generates high-difficulty questions that challenge LLMs' reasoning abilities across various disciplines.

The introduction of InEx presents a novel approach to mitigating hallucinations in large language models (LLMs) by employing a training-free, multi-agent framework that incorporates introspective reasoning and cross-modal collaboration. This method aims to enhance the reliability of multimodal LLMs (MLLMs) by autonomously refining responses through iterative verification processes.

A systematic survey on Deep Research (DR) has been published, highlighting the evolution of large language models (LLMs) from mere text generators to sophisticated problem solvers. This survey outlines a three-stage roadmap for integrating LLMs with external tools, enabling them to tackle complex tasks that require critical thinking and multi-source verification.

A new framework named promptolution has been introduced to optimize prompts for large language models (LLMs), addressing the challenges of existing isolated implementations. This unified, modular open-source system integrates various prompt optimizers, facilitating easier adoption for both researchers and practitioners.

A recent study investigates the alignment between large language models (LLMs) and human brain processes, focusing on how layer-wise representations in LLMs correspond to neural responses during sentence comprehension. By analyzing data from 14 LLMs and fMRI scans of participants listening to a narrative, researchers identified significant correlations between model layers and brain activity.

Large language model-powered chatbots have significantly changed the way individuals access information, particularly in critical areas like mental health. However, their effectiveness in safely managing crises such as suicidal thoughts and self-harm remains uncertain due to the absence of standardized crisis classifications and clinical evaluation methods. This study introduces a taxonomy of crisis categories, a dataset of mental health inputs, and a clinical response assessment protocol to enhance crisis management by LLMs.

The introduction of Fairy2i marks a significant advancement in the field of artificial intelligence, particularly in the quantization of large language models (LLMs). This universal framework enables the transformation of pre-trained real-valued layers into a widely-linear complex form, facilitating extremely low-bit quantization while leveraging existing model checkpoints.

The introduction of STRIDE (Systematic Task Reasoning Intelligence Deployment Evaluator) offers a structured framework for selecting between AI modalities, including direct LLM calls, guided AI assistants, and fully autonomous agentic AI. This framework addresses the complexities and risks associated with deploying agentic AI indiscriminately, ensuring that such autonomy is reserved for tasks requiring dynamic reasoning and evolving contexts.