Search

Multi-agent strategies have emerged as a promising approach to enhance the reasoning abilities of Large Language Models (LLMs) by assigning specialized roles in the problem-solving process. Concurrently, Tree of Thoughts (ToT) methods have shown potential in improving reasoning for complex question-answering tasks by exploring diverse reasoning paths. A critical limitation in multi-agent reasoning is the 'Reasoner' agent's shallow exploration of reasoning paths. While ToT strategies could help mitigate this problem, they may generate flawed reasoning branches, which could harm the trustworthiness of the final answer. To leverage the strengths of both multi-agent reasoning and ToT strategies, we introduce a novel approach combining ToT-based Reasoner agents with a Thought Validator agent. Multiple Reasoner agents operate in parallel, employing ToT to explore diverse reasoning paths. The Thought Validator then scrutinizes these paths, considering a Reasoner's conclusion only if its reasoning is valid. This method enables a more robust voting strategy by discarding faulty reasoning paths, enhancing the system's ability to tackle tasks requiring systematic and trustworthy reasoning. Our method demonstrates superior performance compared to existing techniques when evaluated on the GSM8K dataset, outperforming the standard ToT strategy by an average 5.6% across four LLMs. The code and related content can be found in: https://github.com/SecureAIAutonomyLab/MA-ToT

Recent advancements in AI safety have led to increased efforts in training and red-teaming large language models (LLMs) to mitigate unsafe content generation. However, these safety mechanisms may not be comprehensive, leaving potential vulnerabilities unexplored. This paper introduces MathPrompt, a novel jailbreaking technique that exploits LLMs' advanced capabilities in symbolic mathematics to bypass their safety mechanisms. By encoding harmful natural language prompts into mathematical problems, we demonstrate a critical vulnerability in current AI safety measures. Our experiments across 13 state-of-the-art LLMs reveal an average attack success rate of 73.6\%, highlighting the inability of existing safety training mechanisms to generalize to mathematically encoded inputs. Analysis of embedding vectors shows a substantial semantic shift between original and encoded prompts, helping explain the attack's success. This work emphasizes the importance of a holistic approach to AI safety, calling for expanded red-teaming efforts to develop robust safeguards across all potential input types and their associated risks.

Event detection and text reasoning have become critical applications across various domains. While LLMs have recently demonstrated impressive progress in reasoning abilities, they often struggle with event detection, particularly due to the absence of training methods that consider causal relationships between event triggers and types. To address this challenge, we propose a novel approach for instruction fine-tuning LLMs for event detection. Our method introduces Semantic Causal Graphs (SCGs) to capture both causal relationships and contextual information within text. Building off of SCGs, we propose SCG Instructions for fine-tuning LLMs by focusing on event triggers and their relationships to event types, and employ Low-Rank Adaptation (LoRA) to help preserve the general reasoning abilities of LLMs. Our evaluations demonstrate that training LLMs with SCG Instructions outperforms standard instruction fine-tuning by an average of 35.69\% on Event Trigger Classification. Notably, our fine-tuned Mistral 7B model also outperforms GPT-4 on key event detection metrics by an average of 31.01\% on Event Trigger Identification, 37.40\% on Event Trigger Classification, and 16.43\% on Event Classification. We analyze the retention of general capabilities, observing only a minimal average drop of 2.03 points across six benchmarks. This comprehensive study investigates multiple LLMs for the event detection task across various datasets, prompting strategies, and training approaches.

Software security remains a critical concern, particularly as junior developers, often lacking comprehensive knowledge of security practices, contribute to codebases. While there are tools to help developers proactively write secure code, their actual effectiveness in helping developers fix their vulnerable code remains largely unmeasured. Moreover, these approaches typically focus on classifying and localizing vulnerabilities without highlighting the specific code segments that are the root cause of the issues, a crucial aspect for developers seeking to fix their vulnerable code. To address these challenges, we conducted a comprehensive study evaluating the efficacy of existing methods in helping junior developers secure their code. Our findings across five types of security vulnerabilities revealed that current tools enabled developers to secure only 36.2\% of vulnerable code. Questionnaire results from these participants further indicated that not knowing the code that was the root cause of the vulnerability was one of their primary challenges in repairing the vulnerable code. Informed by these insights, we developed an automated vulnerability root cause (RC) toolkit called T5-RCGCN, that combines T5 language model embeddings with a graph convolutional network (GCN) for vulnerability classification and localization. Additionally, we integrated DeepLiftSHAP to identify the code segments that were the root cause of the vulnerability. We tested T5-RCGCN with 56 junior developers across three datasets, showing a 28.9\% improvement in code security compared to previous methods. Developers using the tool also gained a deeper understanding of vulnerability root causes, resulting in a 17.0\% improvement in their ability to secure code independently. These results demonstrate the tool's potential for both immediate security enhancement and long-term developer skill growth.

Social media platforms are being increasingly used by malicious actors to share unsafe content, such as images depicting sexual activity, cyberbullying, and self-harm. Consequently, major platforms use artificial intelligence (AI) and human moderation to obfuscate such images to make them safer. Two critical needs for obfuscating unsafe images is that an accurate rationale for obfuscating image regions must be provided, and the sensitive regions should be obfuscated (\textit{e.g.} blurring) for users' safety. This process involves addressing two key problems: (1) the reason for obfuscating unsafe images demands the platform to provide an accurate rationale that must be grounded in unsafe image-specific attributes, and (2) the unsafe regions in the image must be minimally obfuscated while still depicting the safe regions. In this work, we address these key issues by first performing visual reasoning by designing a visual reasoning model (VLM) conditioned on pre-trained unsafe image classifiers to provide an accurate rationale grounded in unsafe image attributes, and then proposing a counterfactual explanation algorithm that minimally identifies and obfuscates unsafe regions for safe viewing, by first utilizing an unsafe image classifier attribution matrix to guide segmentation for a more optimal subregion segmentation followed by an informed greedy search to determine the minimum number of subregions required to modify the classifier's output based on attribution score. Extensive experiments on uncurated data from social networks emphasize the efficacy of our proposed method. We make our code available at: https://github.com/SecureAIAutonomyLab/ConditionalVLM

With the recent proliferation of Large Language Models (LLMs), there has been an increasing demand for tools to detect machine-generated text. The effective detection of machine-generated text face two pertinent problems: First, they are severely limited in generalizing against real-world scenarios, where machine-generated text is produced by a variety of generators, including but not limited to GPT-4 and Dolly, and spans diverse domains, ranging from academic manuscripts to social media posts. Second, existing detection methodologies treat texts produced by LLMs through a restrictive binary classification lens, neglecting the nuanced diversity of artifacts generated by different LLMs. In this work, we undertake a systematic study on the detection of machine-generated text in real-world scenarios. We first study the effectiveness of state-of-the-art approaches and find that they are severely limited against text produced by diverse generators and domains in the real world. Furthermore, t-SNE visualizations of the embeddings from a pretrained LLM's encoder show that they cannot reliably distinguish between human and machine-generated text. Based on our findings, we introduce a novel system, T5LLMCipher, for detecting machine-generated text using a pretrained T5 encoder combined with LLM embedding sub-clustering to address the text produced by diverse generators and domains in the real world. We evaluate our approach across 9 machine-generated text systems and 9 domains and find that our approach provides state-of-the-art generalization ability, with an average increase in F1 score on machine-generated text of 19.6\% on unseen generators and domains compared to the top performing existing approaches and correctly attributes the generator of text with an accuracy of 93.6\%.

We describe methods, tools, and a software library called LASPATED, available on GitHub (at https://github.com/vguigues/) to fit models using spatio-temporal data and space-time discretization. A video tutorial for this library is available on YouTube. We consider two types of methods to estimate a non-homogeneous Poisson process in space and time. The methods approximate the arrival intensity function of the Poisson process by discretizing space and time, and estimating arrival intensity as a function of subregion and time interval. With such methods, it is typical that the dimension of the estimator is large relative to the amount of data, and therefore the performance of the estimator can be improved by using additional data. The first method uses additional data to add a regularization term to the likelihood function for calibrating the intensity of the Poisson process. The second method uses additional data to estimate arrival intensity as a function of covariates. We describe a Python package to perform various types of space and time discretization. We also describe two packages for the calibration of the models, one in Matlab and one in C++. We demonstrate the advantages of our methods compared to basic maximum likelihood estimation with simulated and real data. The experiments with real data calibrate models of the arrival process of emergencies to be handled by the Rio de Janeiro emergency medical service.

Calls to improve relationships between research and practice abound, among them efforts to help researchers to work in partnership with and communicate more effectively with policy and practice audiences, informing those that govern and design educational policies and those charged with implementation in schools and classrooms, respectively. These calls range from advocacy for participatory models of research to revisions to academic incentives that presently prioritize contributions to literature above impact on policy or practice. Most recently, the National Academies of Science, Engineering, and Medicine report, "The Future of Education Research at the Institute of Education Sciences in the U.S. Department of Education" (2022), emphasizes that the largest federal funder of education research should include knowledge mobilization among its core areas of work. The end goal of these efforts is to have a more impactful research enterprise, one that both meets the needs of and is taken up in decisions made by policymakers and practitioners. The Center for Research Use in Education (CRUE) seeks to better understand how to strengthen the relationships between research and practice to the same ends--increasing the impact of research by better meeting the needs of policy and practice and by producing work that is useful and used in decision-making. CRUE has approached this work with attention to bidirectionality--focusing on both decision-processes in the practice space and the role of research in them (Farley-Ripple et al., 2022a, 2022b), as well as research production and the strategies used to increase its impact (Van Horn et al, 2023). In this paper the authors describe the work on the latter, presenting results of a mixed methods inquiry into the knowledge mobilization practices of researchers.

To estimate the long-term effects of the Reading Recovery® intervention, a regression discontinuity design (RD) was implemented in a randomly selected sample of Reading Recovery schools during each year of the federally-funded i3 Scale-Up external evaluation (2011-2015) and also in one additional cohort during the 2016-17 school year. Long-term outcomes were measured by collecting scale scores on state achievement tests in reading or English language arts (ELA) that were administered to students during 3rd and 4th grades (2013-2019). We were able to record and link 3rd grade state test score data in reading/ELA for 9,879 students, and 4th grade state test data for 6,345 students. Results suggest that the long-term impact of Reading Recovery on students' reading/ELA test scores in 3rd and 4th grades is statistically significant and substantially negative; students who participated in Reading Recovery in first grade had 3rd and 4th grade state test scores in reading/ELA that were, on average, 0.19 to 0.43 standard deviations (about one-half to one full grade level) below the state test scores of similar students who did not participate in Reading Recovery.

Protein motions occur on multiple time and distance scales. Large-scale motions of protein tertiary-structure elements, i.e. domains, are particularly intriguing as they are essential for the catalytic activity of many enzymes and for the functional cycles of protein machines and motors. Theoretical estimates suggest that domain motions should be very fast, occurring on the nanosecond or microsecond time scales. Indeed, free-energy barriers for domain motions are likely to involve salt bridges, which can break in microseconds. Experimental methods that can directly probe domain motions on fast time scales have appeared only in recent years. This Perspective discusses briefly some of these techniques, including NMR and single-molecule fluorescence spectroscopies. We introduce a few recent studies that demonstrate ultrafast domain motions, and discuss their potential roles. Particularly surprising is the observation of tertiary-structure element dynamics that are much faster than the functional cycles in some protein machines. These swift motions can be rationalized on a case-by-case basis. For example, fast domain closure in multi-substrate enzymes may be utilized to optimize relative substrate orientation. Whether a large mismatch in time scales of conformational dynamics vs. functional cycles is a general design principle in proteins remains to be determined., Comment: Journal of Chemical Physics, accepted Perspective