The field of Artificial Intelligence (AI) is advancing at a rapid rate; specifically, the Large Language Models have become indispensable in modern AI applications. These LLMs have inbuilt safety mechanisms that prevent them from generating unethical and harmful outputs. However, these mechanisms are vulnerable to simple adaptive jailbreaking attacks. The researchers have demonstrated that even… →

Retrieval Augmented Generation is an efficient solution for knowledge-intensive tasks that improves the quality of outputs and makes it more deterministic with minimal hallucinations. However, RAG outputs can still be noisy and may fail to respond appropriately to complex queries. To address this limitation, iterative retrieval updates have been introduced, which update re-retrieval results to… →

The field of robotic manipulation has witnessed a remarkable transformation with the emergence of vision-language-action (VLA) models. These advanced computational frameworks have demonstrated significant potential in executing complex manipulation tasks across diverse environments. Despite their impressive capabilities, VLA models encounter substantial challenges in generalizing across novel contexts, including different objects, environments, and semantic scenarios.  The… →

Integrating vision and language processing in AI has become a cornerstone for developing systems capable of simultaneously understanding visual and textual data, i.e., multimodal data. This interdisciplinary field focuses on enabling machines to interpret images, extract relevant textual information, and discern spatial and contextual relationships. These capabilities promise to reshape real-world applications by bridging the… →

Large language models (LLMs) excel in generating contextually relevant text; however, ensuring compliance with data privacy regulations, such as GDPR, requires a robust ability to unlearn specific information effectively. This capability is critical for addressing privacy concerns where data must be entirely removed from models and any logical connections that could reconstruct deleted information. The… →

Vision-and-language models (VLMs) are important tools that use text to handle different computer vision tasks. Tasks like recognizing images, reading text from images (OCR), and detecting objects can be approached as answering visual questions with text responses. While VLMs have shown limited success on tasks, what remains unclear is how they process and represent multimodal… →

LLMs have revolutionized artificial intelligence with their remarkable scalability and adaptability. Models like GPT-4 and Claude, built with trillions of parameters, demonstrate exceptional performance across diverse tasks. However, their monolithic design comes with significant challenges, including high computational costs, limited flexibility, and difficulties in fine-tuning for domain-specific needs due to risks like catastrophic forgetting and… →

Snowflake recently announced the launch of Arctic Embed L 2.0 and Arctic Embed M 2.0, two small and powerful embedding models tailored for multilingual search and retrieval. The Arctic Embed 2.0 models are available in two distinct variants: medium and large. Based on Alibaba’s GTE-multilingual framework, the medium model incorporates 305 million parameters, of which… →

Language Agents (LAs) have recently become the focal point of research and development because of the significant advancement in large language models (LLMs). LLMs have demonstrated significant advancements in understanding and producing human-like text. LLMs perform various tasks with great performance and accuracy. Through well-designed prompts and carefully selected in-context demonstrations, LLM-based agents, such as… →