Large and diverse datasets have been the cornerstones of many impressive advancements in artificial intelligence. Intelligent creatures, however, learn by interacting with the environment, which changes the input sensory signals and the state of the environment. In this work, we aim to bring the best of both worlds and propose an algorithm that exhibits an exploratory behavior whilst it utilizes large diverse datasets. Our key idea is to leverage deep generative models that are pretrained on static datasets and introduce a dynamic model in the latent space. The transition dynamics simply mixes an action and a random sampled latent. It then applies an exponential moving average for temporal persistency, the resulting latent is decoded to image using pretrained generator. We then employ an unsupervised reinforcement learning algorithm to explore in this environment and perform unsupervised representation learning on the collected data. We further leverage the temporal information of this data to pair data points as a natural supervision for representation learning. Our experiments suggest that the learned representations can be successfully transferred to downstream tasks in both vision and reinforcement learning domains.

Large and diverse datasets have been the cornerstones of many impressive advancements in artificial intelligence. Intelligent creatures, however, learn by interacting with the environment, which changes the input sensory signals and the state of the environment. In this work, we aim to bring the best of both worlds and propose an algorithm that exhibits an exploratory behavior whilst it utilizes large diverse datasets. Our key idea is to leverage deep generative models that are pretrained on static datasets and introduce a dynamic model in the latent space. The transition dynamics simply mixes an action and a random sampled latent.

Large Language Models (LLMs) are increasingly powerful but remain vulnerable to prompt injection attacks, where malicious inputs cause the model to deviate from its intended instructions. This paper introduces Sentinel, a novel detection model, qualifire/prompt-injection-sentinel, based on the \answerdotai/ModernBERT-large architecture. By leveraging ModernBERT's advanced features and fine-tuning on an extensive and diverse dataset comprising a few open-source and private collections, Sentinel achieves state-of-the-art performance. This dataset amalgamates varied attack types, from role-playing and instruction hijacking to attempts to generate biased content, alongside a broad spectrum of benign instructions, with private datasets specifically targeting nuanced error correction and real-world misclassifications. On a comprehensive, unseen internal test set, Sentinel demonstrates an average accuracy of 0.987 and an F1-score of 0.980. Furthermore, when evaluated on public benchmarks, it consistently outperforms strong baselines like protectai/deberta-v3-base-prompt-injection-v2. This work details Sentinel's architecture, its meticulous dataset curation, its training methodology, and a thorough evaluation, highlighting its superior detection capabilities.

Large and diverse datasets have been the cornerstones of many impressive advancements in artificial intelligence. Intelligent creatures, however, learn by interacting with the environment, which changes the input sensory signals and the state of the environment. In this work, we aim to bring the best of both worlds and propose an algorithm that exhibits an exploratory behavior whilst it utilizes large diverse datasets. Our key idea is to leverage deep generative models that are pretrained on static datasets and introduce a dynamic model in the latent space. The transition dynamics simply mixes an action and a random sampled latent.

Phishing has been a prevalent cyber threat that manipulates users into revealing sensitive private information through deceptive tactics, designed to masquerade as trustworthy entities. Over the years, proactively detection of phishing URLs (or websites) has been established as an widely-accepted defense approach. In literature, we often find supervised Machine Learning (ML) models with highly competitive performance for detecting phishing websites based on the extracted features from both phishing and benign (i.e., legitimate) websites. However, it is still unclear if these features or indicators are dependent on a particular dataset or they are generalized for overall phishing detection. In this paper, we delve deeper into this issue by analyzing two publicly available phishing URL datasets, where each dataset has its own set of unique and overlapping features related to URL string and website contents. We want to investigate if overlapping features are similar in nature across datasets and how does the model perform when trained on one dataset and tested on the other. We conduct practical experiments and leverage explainable AI (XAI) methods such as SHAP plots to provide insights into different features' contributions in case of phishing detection to answer our primary question, "Can features for phishing URL detection be trusted across diverse dataset?". Our case study experiment results show that features for phishing URL detection can often be dataset-dependent and thus may not be trusted across different datasets even though they share same set of feature behaviors.

We introduce INSTRUCTOR, a new method for computing text embeddings given task instructions: every text input is embedded together with instructions explaining the use case (e.g., task and domain descriptions). Unlike encoders from prior work that are more specialized, INSTRUCTOR is a single embedder that can generate text embeddings tailored to different downstream tasks and domains, without any further training. We first annotate instructions for 330 diverse tasks and train INSTRUCTOR on this multitask mixture with a contrastive loss. We evaluate INSTRUCTOR on 70 embedding evaluation tasks (66 of which are unseen during training), ranging from classification and information retrieval to semantic textual similarity and text generation evaluation. INSTRUCTOR, while having an order of magnitude fewer parameters than the previous best model, achieves state-of-the-art performance, with an average improvement of 3.4% compared to the previous best results on the 70 diverse datasets. Our analysis suggests that INSTRUCTOR is robust to changes in instructions, and that instruction finetuning mitigates the challenge of training a single model on diverse datasets. Our model, code, and data are available at https://instructor-embedding.github.io.

New research published in PLOS Digital Health is calling attention to disparities in artificial intelligence that could inhibit its ability to be effectively deployed in clinical settings. Researchers analyzed more than 30,000 artificial intelligence clinical papers published in PubMed in 2019 and found that more than 50% of AI studies utilized databases from the U.S. or China, and that almost all the top 10 databases and author nationalities were from high income countries. Such homogenous datasets, the authors explained, can create research bias that hinders the clinical efficacy of AI applications. "The introduction of AI into healthcare comes with its own biases and disparities; it risks thrusting the world toward an exaggerated state of healthcare inequity," William Greig Mitchell, of the Harvard TH Chan School of Public Health in Boston, Massachusetts, and co-authors wrote. "Repeatedly feeding models with relatively homogeneous data, suffering from a lack of diversity in terms of underlying patient populations and often curated from restricted clinical settings, can severely limit the generalizability of results and yield biased AI-based decisions."

Although robots are more than capable today of carrying out all kinds of business tasks efficiently and accurately, the concept of building machines that can think like humans has always been a dream for tech companies and smart city developers. However, the actual way in which the human mind works and processes information is up for debate, with several parties having conflicting opinions regarding the same. Once enough data is generated, simulation models can be created to build software that can think along the same rational or emotional lines as humans. Human thinking is generally influenced by a variety of factors--cognitive, behavioral, geometric, kinematic and physical. Using cognitive modeling, such factors can be considered while attempting to create robots that think and behave like humans. The concept of human thinking is still too vague to be accurately replicated in robots.

A study presented at the Radiological Society of North America (RSNA) 2021 Annual Meeting demonstrates the importance of using racially diverse datasets while training artificial intelligence (AI) systems to ensure fair outcomes. "As the rapid development of deep learning in medicine continues, there are concerns of potential bias when interpreting radiological images," the authors wrote. "As future medical AI systems are approved by regulators, it is crucial that model performance on different racial/ethnic groups is shared to ensure that safe and fair systems are being implemented." The findings were presented by Brandon Price, a medical student at Florida State University College of Medicine in Tallahassee. Many studies have shown that deep learning systems are subjective in their interpretation of data.

Therefore, to make this process better, employers have begun leveraging algorithmic techniques, in order to hire quality candidates. But the question that often comes up is if hiring algorithms prevent bias or amplify it? Algorithmic screening tools, on the surface seem like an appealing replacement for biased human evaluations. However, there experts have started realizing that these tools reproduce and sometimes magnify human biases found in the datasets based on which these tools are designed. Algorithms do not question the human decisions underlying a dataset – they attempt to replicate past decisions, and this can lead them to replicate all the human biases they were intended to remove in the first place.