The Week in AI: Accenture's AI Refinery, DeepMind's AlphaProof and AlphaGeometry 2, More -- Pure AI

The Week in AI: Accenture's AI Refinery, DeepMind's AlphaProof and AlphaGeometry 2, More

By Pure AI Editors
07/29/2024

This edition of our weekly roundup of AI products and services includes Accenture's new AI Refinery framework, DeepMind's new AI models, Pryon's new ingestion engine, More

OpenAI unveiled SearchGPT, a new AI-powered search engine prototype designed to provide more organized and meaningful search results by summarizing and contextualizing information from the web. The service, integrated into ChatGPT, aims to streamline the process of finding relevant information online by enhancing conversational capabilities with real-time data. SearchGPT features a user-friendly interface that prompts users with "What are you looking for?" and delivers concise summaries, descriptions, and attribution links. It also allows users to ask follow-up questions and explore related topics via a sidebar. The service is initially available to around 10,000 test users, with plans for broader integration into ChatGPT. (Our reporting here.)

Accenture launched its new AI Refinery framework, which leverages NVIDIA's AI Foundry to enable businesses to develop custom Large Language Models (LLMs) using the newly introduced Llama 3.1 collection. The framework, which is part of Accenture's foundation model services, provides tools for creating domain-specific LLMs and AI systems that align with the user's business needs. It enables domain model customization, a switchboard platform for model selection, an enterprise cognitive brain for data indexing, and an agentic architecture for autonomous AI system operations. The company is utilizing the AI Refinery internally, starting with marketing and communications, and plans to expand its use to other business functions.

Google DeepMind introduced two innovative AI models designed for advanced mathematical reasoning. The models, AlphaProof and AlphaGeometry 2, were developed to tackle complex math problems that have traditionally posed challenges for existing AI systems. AlphaProof, a reinforcement-learning model, is capable of formal mathematical reasoning; AlphaGeometry 2 is an enhanced version of DeepMind's previous geometry-solving system. The models utilize formal language reasoning and the Gemini large language model to bridge gaps in AI problem-solving capabilities. They were tested using problems from the International Mathematical Olympiad (IMO). They solved four out of six questions. AlphaProof addressed algebra and number theory problems, while AlphaGeometry 2 solved a geometry question.

Pryon launched the Pryon Ingestion Engine and associated SDK, an Extract, Transform, and Load (ETL) solution for converting unstructured data into machine-readable formats for AI applications. The new tool utilizes Pryon's proprietary document processing models to transform data from various sources into structured formats, enhancing the usability and accuracy of AI systems. The Pryon Ingestion Engine and SDK incorporate such advanced techniques as OCR, Handwritten Text Recognition, and semantic segmentation to extract and contextualize data, making it easier for developers to create AI applications, refine large language models, and deploy retrieval-augmented generative solutions. The platform is designed for scalability and performance, capable of handling large volumes of data efficiently, the company says.

OpsVeda, a provider of Operations Command Center (OCC) solutions, announced the availability of Juni, a generative AI-powered virtual operations analyst, on the Microsoft Teams Store. Juni uses natural language processing and domain expertise to answer business queries by accessing real-time data from OpsVeda's intelligence platform, providing timely insights for decision-making. The tool was developed to enhance operational efficiency by allowing frontline team members to make informed decisions without relying on central analyst pools. Juni's capabilities include real-time alerts, data visualization, and direct navigation to relevant OpsVeda dashboards, facilitating rapid response to operational challenges. The integration with Microsoft Teams enables seamless, conversational interaction with Juni, akin to consulting an informed colleague, thereby optimizing demand-supply-inventory-logistics alignment for enterprises.

Hitachi Vantara, a subsidiary of Hitachi, Ltd., announced the general availability of its Hitachi iQ portfolio, which includes AI-ready infrastructure, solutions, and services. The new offerings, including the NVIDIA DGX BasePOD certified infrastructure, are designed to support advanced AI workloads, providing high performance and reliability for critical applications. Hitachi iQ features the Hitachi Content Software for File solution, which accelerates data processing to optimize insights and drive business transformation. The company also launched the AI Discovery Service to assist customers in identifying valuable AI use cases and developing strategic roadmaps. The Hitachi iQ and AI Discovery Service are available globally.

SensiML, a subsidiary of QuickLogic, introduced a new generative AI feature to its Data Studio platform. This enhancement was designed to enable embedded device developers to create synthetic speech datasets using text-to-speech (TTS) and AI voice generation, facilitating the development of advanced speech recognition models for IoT edge devices. The feature, utilizing technology from ElevenLabs, allows the creation of hyper-realistic synthetic speech, tailored to specific applications. This streamlines the process of developing voice-enabled IoT devices, reducing time-to-market by eliminating the need for extensive manual data collection. It supports a wide range of languages and dialects, making it versatile for global applications.

BlackBerry announced CylanceMDR Pro, a new managed detection and response (MDR) service built on its AI-powered Open XDR platform. The service was developed to enhance cybersecurity for businesses by addressing the operational challenges security teams face against increasingly sophisticated cyber threats across expanding attack surfaces. Powered by BlackBerry's Cylance AI, the service analyzes telemetry to detect threats across endpoints, networks, clouds, and more, offering high-fidelity detections and rapid response. It offers an extended detection and response architecture that integrates with existing security tools, providing deep visibility across IT setups without requiring infrastructure overhauls. The service includes over 300 platform integrations, transforming existing data and applications into effective security sensors. CylanceMDR Pro will be available to customers starting July 31.