Mathematical reasoning has proven to be a critical yet challenging task for large language models (LLMs), as they often struggle with complex multi-step problems. To address these limitations, we introduce the Monte Carlo Nash Equilibrium Self-Refine Tree (MC-NEST) algorithm, an enhancement of the Monte Carlo Tree Self-Refine (MCTSr) approach. By integrating Nash Equilibrium strategies with LLM-based self-refinement and self-evaluation processes, MC-NEST aims to improve decision-making for complex mathematical reasoning tasks. This method ensures balanced exploration and exploitation of potential solutions, leveraging Upper Confidence Bound (UCT) scores and various selection policies. Through iterative critique and refinement, MC-NEST enhances the reasoning capabilities of LLMs, particularly for problems requiring strategic decision-making. Comparative analysis reveals that GPT-4o, equipped with MC-NEST using an Importance Sampling Policy, achieved superior accuracy in domains such as Number Theory and Geometry. These results suggest that both LLMs GPT-4o and Phi-3-mini can benefit from MC-NEST, with iterative self-refinement proving especially effective in expanding the reasoning capacity and problem-solving performance of LLMs. We evaluate the effectiveness of MC-NEST on challenging Olympiad-level benchmarks, demonstrating its potential to significantly boost complex mathematical reasoning performance in LLMs.

We present a synthetic data approach for instruction-tuning large language models (LLMs) for low-resource languages in a data-efficient manner, specifically focusing on Thai. We identify three key properties that contribute to the effectiveness of instruction-tuning datasets: fluency, diversity, and cultural context. We propose a seed-data-free framework for generating synthetic instruction-tuning data that incorporates these essential properties. Our framework employs an LLM to generate diverse topics, retrieve relevant contexts from Wikipedia, and create instructions for various tasks, such as question answering, summarization, and conversation. The experimental results show that our best-performing synthetic dataset, which incorporates all three key properties, achieves competitive performance using only 5,000 instructions when compared to state-of-the-art Thai LLMs trained on hundreds of thousands of instructions. Our code and dataset are publicly available at https://github.com/parinzee/seed-free-synthetic-instruct.

"Gold" and "ground truth" human-mediated labels have error. The effects of this error can escape commonly reported metrics of label quality or obscure questions of accuracy, bias, fairness, and usefulness during model evaluation. This study demonstrates methods for answering such questions even in the context of very low reliabilities from expert humans. We analyze human labels, GPT model ratings, and transformer encoder model annotations describing the quality of classroom teaching, an important, expensive, and currently only human task. We answer the question of whether such a task can be automated using two Large Language Model (LLM) architecture families--encoders and GPT decoders, using novel approaches to evaluating label quality across six dimensions: Concordance, Confidence, Validity, Bias, Fairness, and Helpfulness. First, we demonstrate that using standard metrics in the presence of poor labels can mask both label and model quality: the encoder family of models achieve state-of-the-art, even "super-human", results across all classroom annotation tasks. But not all these positive results remain after using more rigorous evaluation measures which reveal spurious correlations and nonrandom racial biases across models and humans. This study then expands these methods to estimate how model use would change to human label quality if models were used in a human-in-the-loop context, finding that the variance captured in GPT model labels would worsen reliabilities for humans influenced by these models. We identify areas where some LLMs, within the generalizability of the current data, could improve the quality of expensive human ratings of classroom instruction.

Code snippet adaptation is a fundamental activity in the software development process. Unlike code generation, code snippet adaptation is not a "free creation", which requires developers to tailor a given code snippet in order to fit specific requirements and the code context. Recently, large language models (LLMs) have confirmed their effectiveness in the code generation task with promising results. However, their performance on adaptation, a reuse-oriented and context-dependent code change prediction task, is still unclear. To bridge this gap, we conduct an empirical study to investigate the performance and issues of LLMs on the adaptation task. We first evaluate the adaptation performances of three popular LLMs and compare them to the code generation task. Our result indicates that their adaptation ability is weaker than generation, with a nearly 15% decrease on pass@1 and more context-related errors. By manually inspecting 200 cases, we further investigate the causes of LLMs' sub-optimal performance, which can be classified into three categories, i.e., Unclear Requirement, Requirement Misalignment and Context Misapplication. Based on the above empirical research, we propose an interactive prompting approach to eliciting LLMs' adaptation ability. Experimental result reveals that our approach greatly improve LLMs' adaptation performance. The best-performing Human-LLM interaction successfully solves 159 out of the 202 identified defects and improves the pass@1 and pass@5 by over 40% compared to the initial instruction-based prompt. Considering human efforts, we suggest multi-agent interaction as a trade-off, which can achieve comparable performance with excellent generalization ability. We deem that our approach could provide methodological assistance for autonomous code snippet reuse and adaptation with LLMs.

The advancement and extensive application of large language models (LLMs) have been remarkable, including their use in scientific research assistance. However, these models often generate scientifically incorrect or unsafe responses, and in some cases, they may encourage users to engage in dangerous behavior. To address this issue in the field of chemistry, we introduce ChemSafetyBench, a benchmark designed to evaluate the accuracy and safety of LLM responses. ChemSafetyBench encompasses three key tasks: querying chemical properties, assessing the legality of chemical uses, and describing synthesis methods, each requiring increasingly deeper chemical knowledge. Our dataset has more than 30K samples across various chemical materials. We incorporate handcrafted templates and advanced jailbreaking scenarios to enhance task diversity. Our automated evaluation framework thoroughly assesses the safety, accuracy, and appropriateness of LLM responses. Extensive experiments with state-of-the-art LLMs reveal notable strengths and critical vulnerabilities, underscoring the need for robust safety measures. ChemSafetyBench aims to be a pivotal tool in developing safer AI technologies in chemistry. Our code and dataset are available at https://github.com/HaochenZhao/SafeAgent4Chem. Warning: this paper contains discussions on the synthesis of controlled chemicals using AI models.

Speech-driven gesture generation using transformer-based generative models represents a rapidly advancing area within virtual human creation. However, existing models face significant challenges due to their quadratic time and space complexities, limiting scalability and efficiency. To address these limitations, we introduce DiM-Gestor, an innovative end-to-end generative model leveraging the Mamba-2 architecture. DiM-Gestor features a dual-component framework: (1) a fuzzy feature extractor and (2) a speech-to-gesture mapping module, both built on the Mamba-2. The fuzzy feature extractor, integrated with a Chinese Pre-trained Model and Mamba-2, autonomously extracts implicit, continuous speech features. These features are synthesized into a unified latent representation and then processed by the speech-to-gesture mapping module. This module employs an Adaptive Layer Normalization (AdaLN)-enhanced Mamba-2 mechanism to uniformly apply transformations across all sequence tokens. This enables precise modeling of the nuanced interplay between speech features and gesture dynamics. We utilize a diffusion model to train and infer diverse gesture outputs. Extensive subjective and objective evaluations conducted on the newly released Chinese Co-Speech Gestures dataset corroborate the efficacy of our proposed model. Compared with Transformer-based architecture, the assessments reveal that our approach delivers competitive results and significantly reduces memory usage, approximately 2.4 times, and enhances inference speeds by 2 to 4 times. Additionally, we released the CCG dataset, a Chinese Co-Speech Gestures dataset, comprising 15.97 hours (six styles across five scenarios) of 3D full-body skeleton gesture motion performed by professional Chinese TV broadcasters.

Large Audio Language Models (LALM) combine the audio perception models and the Large Language Models (LLM) and show a remarkable ability to reason about the input audio, infer the meaning, and understand the intent. However, these systems rely on Transformers which scale quadratically with the input sequence lengths which poses computational challenges in deploying these systems in memory and time-constrained scenarios. Recently, the state-space models (SSMs) have emerged as an alternative to transformer networks. While there have been successful attempts to replace transformer-based audio perception models with state-space ones, state-space-based LALMs remain unexplored. First, we begin by replacing the transformer-based audio perception module and then replace the transformer-based LLM and propose the first state-space-based LALM. Experimental results demonstrate that space-based LALM despite having a significantly lower number of parameters performs competitively with transformer-based LALMs on close-ended tasks on a variety of datasets.

Although large language models (LLMs) have achieved remarkable success across various domains, their considerable scale necessitates substantial computational resources, posing significant challenges for deployment in resource-constrained environments. Layer pruning, as a simple yet effective compression method, removes layers of a model directly, reducing computational overhead. However, what are the best practices for layer pruning in LLMs? Are sophisticated layer selection metrics truly effective? Does the LoRA (Low-Rank Approximation) family, widely regarded as a leading method for pruned model fine-tuning, truly meet expectations when applied to post-pruning fine-tuning? To answer these questions, we dedicate thousands of GPU hours to benchmarking layer pruning in LLMs and gaining insights across multiple dimensions. Our results demonstrate that a simple approach, i.e., pruning the final 25\% of layers followed by fine-tuning the \texttt{lm\_head} and the remaining last three layer, yields remarkably strong performance. Following this guide, we prune Llama-3.1-8B-It and obtain a model that outperforms many popular LLMs of similar size, such as ChatGLM2-6B, Vicuna-7B-v1.5, Qwen1.5-7B and Baichuan2-7B. We release the optimal model weights on Huggingface, and the code is available on GitHub.

When better to hold a conference on artificial intelligence and the countless ways it is advancing science than in those brief days between the first Nobel prizes being awarded in the field and the winners heading to Stockholm for the lavish white tie ceremony? It was fortuitous timing for Google DeepMind and the Royal Society who this week convened the AI for Science Forum in London. Last month, Google DeepMind bagged the Nobel prize in chemistry a day after AI took the physics prize. Scientists have worked with AI for years, but the latest generation of algorithms have brought us to brink of transformation, Demis Hassabis, the chief executive officer of Google DeepMind, told the meeting. "If we get it right, it should be an incredible new era of discovery and a new golden age, maybe even a kind of new renaissance," he said.

OpenAI has lifted the lid (just a crack) on its safety-testing processes. It has put out two papers describing how it stress-tests its powerful large language models to try to identify potential harmful or otherwise unwanted behavior, an approach known as red-teaming. The first paper describes how OpenAI directs an extensive network of human testers outside the company to vet the behavior of its models before they are released. The second presents a new way to automate parts of the testing process, using a large language model like GPT-4 to come up with novel ways to bypass its own guardrails. MIT Technology Review got an exclusive preview of the work.