News
The Week in AI: Neural Magic's GuideLLM, IBM's Docling, Rune Labs' StrivePD-AI, More
- By Pure AI Editors
- 09/09/2024
This edition of our weekly roundup of AI products and services includes Neural Magic's new open-source GuideLLM, IBM Research's new Docling PDF conversion tool, Rune Labs' StrivePD-AI for improved clinical reporting on Parkinson's patients , Westinghouse Nuclear's Hive System, and more.
Neural Magic unveiled GuideLLM, an open-source tool designed to evaluate and optimize the deployment of large language models (LLMs) in production environments. GuideLLM aims to provide real-time performance insights, improve resource utilization, and offer cost-efficient strategies for organizations deploying LLMs. GuideLLM's key features include performance benchmarking, hardware resource optimization, and cost estimation, making it an essential tool for ensuring LLM deployments meet real-world demands. It also allows users to simulate scalability scenarios, ensuring their applications handle varying traffic loads without compromising performance. With support for Linux and macOS, and installation via Python’s PyPI, GuideLLM is accessible to a wide range of users, from research teams to enterprise operations.
LG AI Research open-sourced EXAONE 3.0, a bilingual large language model (LLM) designed for English and Korean. With 7.8 billion parameters, EXAONE 3.0 excels in real-world applications and complex reasoning tasks, offering top-tier performance across several industry benchmarks, the company said. EXAONE 3.0’s development represents a major leap from earlier versions, with significant reductions in processing time and costs. Its architecture, based on a decoder-only transformer model, is optimized for handling linguistic tasks in both languages. This open-source release is expected to foster global collaboration and innovation within the AI community, aligning with LG's vision of democratizing AI access, the company said.
The AI4K Group at IBM Research launched Docling, an open-source tool for converting complex PDF documents into machine-readable formats. Docling distinguishes itself with AI models specifically designed for layout analysis and table recognition, such as DocLayNet and TableFormer. The tool operates efficiently on commodity hardware and offers both batch processing and interactive options, making it suitable for researchers and businesses alike, the group said. Docling's AI-powered pipeline processes PDFs by parsing text, analyzing layout, and reconstructing tables with high accuracy. It demonstrated robust performance in tests, processing up to 2.45 pages per second on a MacBook Pro and maintaining efficiency even on older hardware, the group said.
Rune Labs, a precision neurology software company, announced the general availability of StrivePD-AI, a new feature within its StrivePD app designed to leverage GenAI to provide monthly clinical reports for Parkinson’s disease patients. These AI-driven reports offer personalized insights into disease progression, symptom trends, and medication efficacy, helping patients, caregivers, and clinicians optimize care between visits, the company said. The reports use data from patient inputs, such as symptoms and activity logs, and are processed through Rune Labs' algorithms, incorporating Anthropic's Claude 3.5 Sonnet LLM. The goal is to empower Parkinson’s patients and their support teams with detailed, actionable information to improve health outcomes.
Hewlett Packard Enterprise (HPE) announced the general availability of HPE Private Cloud AI, a platform designed to streamline the deployment of GenAI applications. Developed in collaboration with NVIDIA, the platform enables organizations to launch virtual AI assistants using private data with just one click, the company said. The new platform includes solution accelerators that automate the process of deploying AI applications, such as virtual assistants for tech support, sales, and marketing. These applications, powered by open-source large language models (LLMs), can be customized for various business needs. HPE also introduced its Unleash AI partner program, connecting users with software vendors, integrators, and service providers to support AI use cases and accelerate deployment across industries like healthcare, retail, and finance.
Voice AI company SoundHound launched new customization tools for its SoundHound Chat AI voice assistant, designed to enhance how automotive brands interact with drivers. The tools, currently being piloted with SoundHound's OEM partners, allow automakers to personalize in-car voice experiences and strengthen their brand identity. The new features include custom Brand ID, enabling automakers to adjust responses to reflect their image, handle sensitive topics, and deploy updates seasonally. Additionally, a new Onboarding and Education tool will help drivers better utilize voice assistant functionalities, offering personalized greetings and guidance. The updates follow SoundHound Chat AI’s integration into several major automotive brands, including Alfa Romeo, Citroën, and DS Automobiles.
AtScale announced the private preview of new Natural Language Query (NQL) capabilities in its namesake universal semantic layer. The new feature allows business users to access data insights by asking questions in plain English, eliminating the need for complex SQL commands or data engineering. Powered by the company's Semantic Layer and Generative AI, NQL provides accurate insights directly from BI reports or custom queries, the company said. The NQL feature, now available to select enterprises, was developed to streamline data analysis by bridging natural language queries with complex database schemas, ultimately accelerating decision-making and reducing costs.
Knoa Software, a provider of task mining and user analytics for enterprise applications, announced two major updates to its task mining solution available in the SAP Store. The enhancements include support for SAP Signavio's AI-assisted process analyzer, currently in beta, and monitoring for non-SAP applications, broadening its existing coverage. These features provide SAP users with deeper insights into process execution, the company said, allowing them to drive improvements across both SAP and non-SAP environments. The AI-powered capabilities, leveraging Knoa and SAP Signavio integration will be fully available by Q4 2024.
Westinghouse Electric Company launched its Hive System, a nuclear-specific GenAI system designed to deliver custom GenAI solutions for its global customer base. Its list of capabilities includes a cutting-edge bertha Nuclear Large Language Model AI System, which named in honor of Bertha Lamme, the first woman in the U.S. to receive a mechanical engineering degree and the first female engineer hired by Westinghouse. The bertha System includes tuned and trained models, methods and tools that can be applied to many different use cases, the company said. By integrating the Hive System into its own products, services, and processes, Westinghouse engineers drive enhancements to their operations and customer applications. They system also helps customers optimize maintenance planning, enhance inspections, and improve the digital user experience to provide operational teams with the right information at the right moment, the company said.