Large language model
AI advancements aim to improve accessibility and usefulness across various communities, ensuring it addresses diverse needs and offers solutions that enhance daily life for all individuals.
ETH Zurich researchers developed an approach using Fast Feedforward Networks (FFF) to increase the speed of Large Language Models (LLM). By engaging only a small fraction of neurons for individual inferences, their UltraFastBERT model could potentially run 341x faster, although a software workaround currently yields a 78x improvement.
Elon Musk’s AI startup, X.AI, is seeking to raise $1 billion through an equity offering after securing $135 million in funding since July. The company aims to advance AI and compete with major players like OpenAI and Google. Their unique chatbot Grok features a distinct personality, drawing on talent from AI leaders for development.
Noah Gift switched his Duke University coding class from Python to the more challenging Rust language, leveraging GitHub’s AI tool Copilot to assist students. Copilot, developed from OpenAI’s GPT-3.5 and GPT-4 models, offers real-time coding assistance. While it’s transforming coding practices and enabling faster code production, there are concerns over IP security and potential quality…
This article details the integration of Large Language Models (LLMs), specifically the “Flan T5” model, with Apache Spark for text data transformations such as sentiment analysis. It provides instructions on setting up Apache Spark and Python, installing necessary libraries, and writing code to create a Spark User-Defined Function (UDF) for sentiment analysis on a dataset.…
This tutorial provides an end-to-end guide on implementing object detection using KerasCV, specifically RetinaNet, to identify healthy and diseased plant leaves. The process involves inspecting and preprocessing data, setting up RetinaNet with a YOLOv8 backbone, training the model with focal loss and smooth L1 loss, and making predictions, considering class imbalance with focal loss. It…
This article explores various methods of matrix multiplication on the M2 MacBook using Go and Metal, including cgo and Metal Shading Language, concluding that GPU-based methods and Metal Performance Shaders are remarkably faster than CPU-based implementations. Benchmarks and GPU usage data support the performance advantages of these GPU-accelerated approaches over Go and OpenBLAS.
University of Geneva researchers have developed Graph Neural Networks (GNN) to predict healthcare-associated infections, outperforming traditional models in early detection of multidrug-resistant Enterobacteriaceae colonization with over 88% accuracy. The GNN model utilizes patient and healthcare worker network data to significantly enhance infection prevention techniques in healthcare settings.
Researchers from the Shanghai AI Lab and MIT have presented the Hierarchically Gated Recurrent Neural Network (HGRN) for efficient sequence modeling. The HGRN integrates forget gates to better handle long-term dependencies in tasks like language modeling and image classification. It surpasses traditional RNNs and Transformers by balancing training efficiency and sequence complexity, with promising results…
Researchers from The Hong Kong University of Science and Technology and Sun Yat-sen University have developed Photo-SLAM, an innovative framework for real-time localization and photorealistic mapping with RGB-D, stereo, and monocular cameras. Photo-SLAM addresses scalability and operational limitations of existing methods and achieves high-fidelity scene rendering at up to 1000 fps. It utilizes Gaussian Pyramid…
The study addresses local private mean estimation of high-dimensional vectors, noting sub-optimal error or high complexity in existing solutions. A new framework, ProjUnit, is proposed, which offers computationally efficient algorithms with low communication complexity and near-optimal error by projecting inputs to a random low-dimensional subspace before normalization.
A study found that observing soft robots assisting with tasks alleviated viewers’ safety worries and job security fears, suggesting a psychological edge over traditional hard-material robots.
A University of Geneva study, led by Alexandre Pouget, demonstrated a machine-learning algorithm can identify Bordeaux red wines’ chateaux of origin by their chemical profiles with 100% accuracy. The algorithm also recognized vintage years with 50% accuracy and confirmed the chemical foundation of terroir.
The text discusses integrating Amazon Comprehend and Amazon Kendra to enrich enterprise search capabilities. Structured and unstructured data are rapidly growing, and using custom metadata helps categorize information. Amazon Comprehend can identify document types and entities, which Amazon Kendra then uses to filter search results, including facets for better searching. The solution is particularly applied…
Stability AI’s SDXL Turbo utilizes Adversarial Diffusion Distillation (ADD) for rapid, high-fidelity text-to-image synthesis, outperforming multi-step models with a single-step process, detailed in their research paper. It’s demonstrated in real-time on Clipdrop and hailed for its exceptional image quality and speed on computational platforms.
Research from various institutions proposes the GREAT PLEA ethical framework for generative AI in healthcare, mirroring military ethics, to ensure transparency, fairness, and empathy in AI deployment, and calls for user education on AI systems to improve trust and patient care.
Small Language Models (SLMs) are emerging as an efficient, adaptable, and secure alternative to Large Language Models, offering benefits in training cost, deployment, transparency, and accuracy for resource-constrained applications. SLMs like DistilBERT, Orca 2, and versions of BERT are increasingly applied in customer service, product development, email automation, and personalized marketing.
Protopia AI and AWS have partnered to provide a tool called Stained Glass Transform (SGT), enabling businesses to deploy large language models (LLMs) securely without compromising data privacy. SGT protects sensitive information in prompts and fine-tuning data by converting them into randomized re-representations, preserving usability and accuracy. This facilitates responsible AI implementation and competitive advantage…
Lasso Security discovered 1,681 exposed API tokens with varying access levels in code on HuggingFace and GitHub, posing significant security risks. Tokens could potentially allow unauthorized modifications to popular AI models, with consequences if misused. The issue was addressed by revoking the compromised tokens.
AstraZeneca invests $247 million in Absci to develop an AI-generated antibody for unspecified cancer treatment. Absci’s AI platform aims to accelerate discovery by simulating protein interactions and validation in wet-labs, potentially revolutionizing oncology drug development with a promised rapid advancement cycle.