We address the challenge of generating diverse attack prompts for large language models (LLMs) that elicit harmful behaviors (e.g., insults, sexual content) and are used for safety fine-tuning. Rather than relying on manual prompt engineering, attacker LLMs can be trained with reinforcement learning (RL) to automatically generate such prompts using only a toxicity classifier as a reward. However, capturing a wide range of harmful behaviors is a significant challenge that requires explicit diversity objectives. Existing diversity-seeking RL methods often collapse to limited modes: once high-reward prompts are found, exploration of new regions is discouraged. Inspired by the active learning paradigm that encourages adaptive exploration, we introduce \textit{Active Attacks}, a novel RL-based red-teaming algorithm that adapts its attacks as the victim evolves. By periodically safety fine-tuning the victim LLM with collected attack prompts, rewards in exploited regions diminish, which forces the attacker to seek unexplored vulnerabilities. This process naturally induces an easy-to-hard exploration curriculum, where the attacker progresses beyond easy modes toward increasingly difficult ones. As a result, Active Attacks uncovers a wide range of local attack modes step by step, and their combination achieves wide coverage of the multi-mode distribution. Active Attacks, a simple plug-and-play module that seamlessly integrates into existing RL objectives, unexpectedly outperformed prior RL-based methods -- including GFlowNets, PPO, and REINFORCE -- by improving cross-attack success rates against GFlowNets, the previous state-of-the-art, from 0.07% to 31.28% (a relative gain greater than $400\ \times$) with only a 6% increase in computation. Our code is publicly available \href{https://github.com/dbsxodud-11/active_attacks}{here}.

Artificial intelligence has been applied in various aspects of online education to facilitate teaching and learning. However, few approaches has been made toward a complete AI-powered tutoring system. In this work, we explore the development of a full-fledged intelligent tutoring system powered by state-of-the-art large language models (LLMs), covering automatic course planning and adjusting, tailored instruction, and flexible quiz evaluation. To make the system robust to prolonged interaction and cater to individualized education, the system is decomposed into three inter-connected core processes-interaction, reflection, and reaction. Each process is implemented by chaining LLM-powered tools along with dynamically updated memory modules. Tools are LLMs prompted to execute one specific task at a time, while memories are data storage that gets updated during education process. Statistical results from learning logs demonstrate the effectiveness and mechanism of each tool usage. Subjective feedback from human users reveal the usability of each function, and comparison with ablation systems further testify the benefits of the designed processes in long-term interaction.

This article explores the transformative impact of artificial intelligence (AI) on scientific research. It highlights ten ways in which AI is revolutionizing the work of scientists, including powerful referencing tools, improved understanding of research problems, enhanced research question generation, optimized research design, stub data generation, data transformation, advanced data analysis, and AI-assisted reporting. While AI offers numerous benefits, challenges such as bias, privacy concerns, and the need for human-AI collaboration must be considered. The article emphasizes that AI can augment human creativity in science but not replace it.

The results of high-resolution machine learning-based estimation of daily soil moisture for the western and midwestern United States are eligible for PeerJ submission. The article is written in high-quality English and is well structured. A consistent and logical presentation of the material contributes to the reader's perception of information. Also, the article is well filled with illustrative material that enhances its informativeness. The tested research algorithm made it possible to qualitatively reveal the essence of the problem of soil moisture assessment using machine learning methods.

Qualitative research is an approach to understanding social phenomenon based around human interpretation of data, particularly text. Probabilistic topic modelling is a machine learning approach that is also based around the analysis of text and often is used to in order to understand social phenomena. Both of these approaches aim to extract important themes or topics in a textual corpus and therefore we may see them as analogous to each other. However there are also considerable differences in how the two approaches function. One is a highly human interpretive process, the other is automated and statistical. In this paper we use this analogy as the basis for our Theme and Topic system, a tool for qualitative researchers to conduct textual research that integrates topic modelling into an accessible interface. This is an example of a more general approach to the design of interactive machine learning systems in which existing human professional processes can be used as the model for processes involving machine learning. This has the particular benefit of providing a familiar approach to existing professionals, that may can make machine learning seem less alien and easier to learn. Our design approach has two elements. We first investigate the steps professionals go through when performing tasks and design a workflow for Theme and Topic that integrates machine learning. We then designed interfaces for topic modelling in which familiar concepts from qualitative research are mapped onto machine learning concepts. This makes these the machine learning concepts more familiar and easier to learn for qualitative researchers.

Artificial intelligence (AI) is emerging as a valuable tool for food and beverage makers looking to bolster front-end innovation. Manufacturers, restaurants, ingredient suppliers, flavor houses and more are leveraging insights from machine learning to get closer to consumer trends and market more nuanced propositions. New food and flavor concepts traditionally have been ascribed to culinary experts, chefs and product developers, said Ron Harnik, vice president of marketing at Tastewise, an Israel-based AI food and beverage platform. Translating an idea into a finished product can take months or even years. "The processes that are set up to take products to market simply aren't built to be quick and accurate enough to reflect how fast consumers are changing," Mr. Harnik said.

The number of variables is virtually numberless, all being studied in both functional roles, like vector-valued functions or y f(x) or x g (y), or both causal directions, direct and reverse. In an undirected simple graph of order n, the maximum degree of each vertex is n 1 and the maximum size of the graph is n(n 1)/2, and n(n 1)/2, if to include loops, what is responsible for the network effects.

Pattern discovery, the process of discovering previously unrecognized patterns, is often performed as an ad-hoc process with little resulting certainty in the quality of the proposed patterns. Pattern validation, the process of validating the accuracy of proposed patterns, remains dominated by the simple heuristic of "the rule of three". This article shows how to use established scientific research methods for the purpose of pattern discovery and validation. We present a specific approach, called the handbook method, that uses the qualitative survey, action research, and case study research for pattern discovery and evaluation, and we discuss the underlying principle of using scientific methods in general. We evaluate the handbook method using three exploratory studies and demonstrate its usefulness.

The idea of using Artificial Intelligence to understand consumer behavior has been around for a while now. From speculating its accuracy to debating its methods, researchers have always had a keen interest in discussing AI's future in consumer research. Emotion AI is the most noticeable development in this regard. Emotion AI is the sub-set of Artificial Intelligence that tries to understand human expressions, both verbal and non-verbal. Also known as Affective Computing, Emotion AI is the science of recognizing, interpreting, processing, and simulating human expressions. Affective Computing was first coined in 1995 by Rosalind Picard's paper of the same name, published by the MIT Press1.

Her research has uncovered racial and gender bias in facial analysis tools sold by companies such as Amazon that have a hard time recognizing certain faces, especially darker-skinned women. Buolamwini holds a white mask she had to use so that software could detect her face. Facial recognition technology was already seeping into everyday life -- from your photos on Facebook to police scans of mugshots -- when Joy Buolamwini noticed a serious glitch: Some of the software couldn't detect dark-skinned faces like hers. That revelation sparked the Massachusetts Institute of Technology researcher to launch a project that's having an outsize influence on the debate over how artificial intelligence should be deployed in the real world. Her tests on software created by brand-name tech firms such as Amazon uncovered much higher error rates in classifying the gender of darker-skinned women than for lighter-skinned men.