Existing Large Multimodal Models (LMMs) struggle with mathematical geometric reasoning due to a lack of high-quality image-text paired data. Current geometric data generation approaches, which apply preset templates to generate geometric data or use Large Language Models (LLMs) to rephrase questions and answers (Q&A), unavoidably limit data accuracy and diversity. To synthesize higherquality data, we propose a two-stage Reverse Chain-of-Thought (R-CoT) geometry problem generation pipeline. First, we introduce GeoChain to produce highfidelity geometric images and corresponding descriptions highlighting relations among geometric elements. We then design a Reverse A&Q method that reasons step-by-step based on the descriptions and generates questions in reverse from the reasoning results. Experiments demonstrate that the proposed method brings significant and consistent improvements on multiple LMM baselines, achieving new performance records in the 2B, 7B, and 8B settings. Notably, R-CoT-8B significantly outperforms previous state-of-the-art open-source mathematical models by 16.6% on MathVista and 9.2% on GeoQA, while also surpassing the closedsource model GPT-4o by an average of 13% across both datasets. The code is available at https://github.com/dle666/R-CoT. Large Language Models (LLMs) exhibit excellent reasoning capabilities and draw extensive attention from the artificial intelligence research community (Lu et al., 2023b) to mathematical problemsolving in textual form (Chen et al., 2024b; Liao et al., 2024; Zhou et al., 2024; Zhao et al., 2024b; Zhou & Zhao, 2024; Kim et al., 2024). However, LLMs still struggle to solve mathematical problems involving images that require visual comprehension. Geometry problems, as typical mathematical problems with images, play an important role in evaluating mathematical reasoning skills (Zhang et al., 2023c), requiring a high level of visual comprehension. Besides, even though some problems are not related to geometry on the surface, they require the same skills for models (e.g., fine-grained image comprehension skills and multi-step reasoning skills). With the appearance of o1 (OpenAI, 2024), GPT-4o (Islam & Moushi, 2024), Gemini (Team et al., 2023), and numerous Large Multimodal Models (LMMs) (Li et al., 2024a; Liu et al., 2024; Chen et al., 2024d; Bai et al., 2023), recent researches progressively investigate using LMMs to solve mathematical geometry problems. Although LMMs show impressive results in general visual question-answering (VQA) tasks (Fan et al., 2024; Liu et al., 2024), they still face challenges in solving mathematical geometry problems. Adjust values in the question and generate answers.

Low-rank adaption (LoRA) is a widely used parameter-efficient finetuning method for LLM that reduces memory requirements. However, current LoRA optimizers lack transformation invariance, meaning the actual updates to the weights depends on how the two LoRA factors are scaled or rotated. This deficiency leads to inefficient learning and sub-optimal solutions in practice. This paper introduces LoRA-RITE, a novel adaptive matrix preconditioning method for LoRA optimization, which can achieve transformation invariance and remain computationally efficient. We provide theoretical analysis to demonstrate the benefit of our method and conduct experiments on various LLM tasks with different models including Gemma 2B, 7B, and mT5-XXL. The results demonstrate consistent improvements against existing optimizers. For example, replacing Adam with LoRA-RITE during LoRA fine-tuning of Gemma-2B yielded 4.6\% accuracy gain on Super-Natural Instructions and 3.5\% accuracy gain across other four LLM benchmarks (HellaSwag, ArcChallenge, GSM8K, OpenBookQA).

The retrieval-augmented generation (RAG) approach is used to reduce the confabulation of large language models (LLMs) for question answering by retrieving and providing additional context coming from external knowledge sources (e.g., by adding the context to the prompt). However, injecting incorrect information can mislead the LLM to generate an incorrect answer. In this paper, we evaluate the effectiveness and robustness of four LLMs against misinformation - Gemma 2, GPT-4o-mini, Llama~3.1, and Mixtral - in answering biomedical questions. We assess the answer accuracy on yes-no and free-form questions in three scenarios: vanilla LLM answers (no context is provided), "perfect" augmented generation (correct context is provided), and prompt-injection attacks (incorrect context is provided). Our results show that Llama 3.1 (70B parameters) achieves the highest accuracy in both vanilla (0.651) and "perfect" RAG (0.802) scenarios. However, the accuracy gap between the models almost disappears with "perfect" RAG, suggesting its potential to mitigate the LLM's size-related effectiveness differences. We further evaluate the ability of the LLMs to generate malicious context on one hand and the LLM's robustness against prompt-injection attacks on the other hand, using metrics such as attack success rate (ASR), accuracy under attack, and accuracy drop. As adversaries, we use the same four LLMs (Gemma 2, GPT-4o-mini, Llama 3.1, and Mixtral) to generate incorrect context that is injected in the target model's prompt. Interestingly, Llama is shown to be the most effective adversary, causing accuracy drops of up to 0.48 for vanilla answers and 0.63 for "perfect" RAG across target models. Our analysis reveals that robustness rankings vary depending on the evaluation measure, highlighting the complexity of assessing LLM resilience to adversarial attacks.

Recent advancements in Large Language Models (LLMs) like ChatGPT and GPT-4 have shown remarkable abilities in a wide range of tasks such as summarizing texts and assisting in coding. Scientific research has demonstrated that these models can also play text-adventure games. This study aims to explore whether LLMs can autonomously create text-based games based on anthropological classics, evaluating also their effectiveness in communicating knowledge. To achieve this, the study engaged anthropologists in discussions to gather their expectations and design inputs for an anthropologically themed game. Through iterative processes following the established HCI principle of 'design thinking', the prompts and the conceptual framework for crafting these games were refined. Leveraging GPT3.5, the study created three prototypes of games centered around the seminal anthropological work of the social anthropologist's Bronislaw Malinowski's "Argonauts of the Western Pacific" (1922). Subsequently, evaluations were conducted by inviting senior anthropologists to playtest these games, and based on their inputs, the game designs were refined. The tests revealed promising outcomes but also highlighted key challenges: the models encountered difficulties in providing in-depth thematic understandings, showed suspectibility to misinformation, tended towards monotonic responses after an extended period of play, and struggled to offer detailed biographical information. Despite these limitations, the study's findings open up new research avenues at the crossroads of artificial intelligence, machine learning, LLMs, ethnography, anthropology and human-computer interaction.

Recently, rapid advancements in Multi-Modal In-Context Learning (MM-ICL) have achieved notable success, which is capable of achieving superior performance across various tasks without requiring additional parameter tuning. However, the underlying rules for the effectiveness of MM-ICL remain under-explored. To fill this gap, this work aims to investigate the research question: "What factors affect the performance of MM-ICL?'' To this end, we investigate extensive experiments on the three core steps of MM-ICL including demonstration retrieval, demonstration ordering, and prompt construction using 6 vision large language models and 20 strategies. Our findings highlight (1) the necessity of a multi-modal retriever for demonstration retrieval, (2) the importance of intra-demonstration ordering over inter-demonstration ordering, and (3) the enhancement of task comprehension through introductory instructions in prompts. We hope this study can serve as a foundational guide for optimizing MM-ICL strategies in future research.

Large language models are probabilistic models, and the process of generating content is essentially sampling from the output distribution of the language model. Existing watermarking techniques inject watermarks into the generated content without altering the output quality. On the other hand, existing acceleration techniques, specifically speculative sampling, leverage a draft model to speed up the sampling process while preserving the output distribution. However, there is no known method to simultaneously accelerate the sampling process and inject watermarks into the generated content. In this paper, we investigate this direction and find that the integration of watermarking and acceleration is non-trivial. We prove a no-go theorem, which states that it is impossible to simultaneously maintain the highest watermark strength and the highest sampling efficiency. Furthermore, we propose two methods that maintain either the sampling efficiency or the watermark strength, but not both. Our work provides a rigorous theoretical foundation for understanding the inherent trade-off between watermark strength and sampling efficiency in accelerating the generation of watermarked tokens for large language models. We also conduct numerical experiments to validate our theoretical findings and demonstrate the effectiveness of the proposed methods.

Watermarking is a technique that involves embedding nearly unnoticeable statistical signals within generated content to help trace its source. This work focuses on a scenario where an untrusted third-party user sends prompts to a trusted language model (LLM) provider, who then generates a text from their LLM with a watermark. This setup makes it possible for a detector to later identify the source of the text if the user publishes it. The user can modify the generated text by substitutions, insertions, or deletions. Our objective is to develop a statistical method to detect if a published text is LLM-generated from the perspective of a detector. We further propose a methodology to segment the published text into watermarked and non-watermarked sub-strings. The proposed approach is built upon randomization tests and change point detection techniques. We demonstrate that our method ensures Type I and Type II error control and can accurately identify watermarked sub-strings by finding the corresponding change point locations. To validate our technique, we apply it to texts generated by several language models with prompts extracted from Google's C4 dataset and obtain encouraging numerical results. We release all code publicly at https://github.com/doccstat/llm-watermark-cpd.

Scientific innovation is pivotal for humanity, and harnessing large language models (LLMs) to generate research ideas could transform discovery. However, existing LLMs often produce simplistic and repetitive suggestions due to their limited ability in acquiring external knowledge for innovation. To address this problem, we introduce an enhanced planning and search methodology designed to boost the creative potential of LLM-based systems. Our approach involves an iterative process to purposely plan the retrieval of external knowledge, progressively enriching the idea generation with broader and deeper insights. Validation through automated and human assessments indicates that our framework substantially elevates the quality of generated ideas, particularly in novelty and diversity. The number of unique novel ideas produced by our framework is 3.4 times higher than without it. Moreover, our method outperforms the current state-of-the-art, generating at least 2.5 times more top-rated ideas based on 170 seed papers in a Swiss Tournament evaluation.

Large Language Models (LLMs) have shown great potential in automating code generation; however, their ability to generate accurate circuit-level SPICE code remains limited due to a lack of hardware-specific knowledge. In this paper, we analyze and identify the typical limitations of existing LLMs in SPICE code generation. To address these limitations, we present SPICEPilot a novel Python-based dataset generated using PySpice, along with its accompanying framework. This marks a significant step forward in automating SPICE code generation across various circuit configurations. Our framework automates the creation of SPICE simulation scripts, introduces standardized benchmarking metrics to evaluate LLM's ability for circuit generation, and outlines a roadmap for integrating LLMs into the hardware design process. SPICEPilot is open-sourced under the permissive MIT license at https://github.com/ACADLab/SPICEPilot.git.

One of the major challenges in training SAEs is the substantial storage and throughput required for latent activations. While text data requires only 2 bytes per token, latent activations occupy 8K bytes per token--resulting in a 4,096x increase in both storage needs and disk throughput. This, combined with the relatively fast training steps of shallow SAEs, means that data loading quickly becomes the main bottleneck in the training process. Due to these infrastructure constraints, we do not save activations in advance but instead generate them on-the-fly. This contrasts with the approach taken by Lieberum et al. (2024); Templeton et al. (2024b), where activations are pre-saved and a high-speed dataloading pipeline is built to keep up with training. To manage this, we adopt a producer-consumer model. Language Models (LMs) generate activations and store them in an activation buffer, while the SAEs consume the activations in random order. The process is serialized: once the buffer is full, SAE training begins, and when half the buffer is consumed, the LMs refill it. Each time the buffer is refilled, we shuffle it to introduce randomness into the training data without needing to save and shuffle all activations at once.