The Week in AI: New Tools from Oracle, Red Hat, Pytorch, Google's DeepMind, More

This edition of our weekly roundup of AI products and services includes new features in Red Hat's OpenShift AI, Pytorch's new ExecuTorch Alpha release, a new version of AlphaFold from Google's DeepMind, SpiNNcloud, German neuromorphic supercomputing company SpiNNcloud Systems' release of SpiNNaker2, and product news from ¬†Oracle, Meta, and Protect AI.

Red Hat announced Red Hat announced new and enhanced features in its Red Hat OpenShift AI, an open hybrid AI and machine learning (ML) platform built on Red Hat OpenShift. The updates were designed to allow enterprises to create and deliver AI-enabled apps at scale across hybrid clouds. They provide access to the latest AI/ML innovations and support from an expansive AI-centric partner ecosystem. The latest version of the platform, Red Hat OpenShift AI 2.9, delivers: model serving at the edge, enhanced model serving, distributed workloads with ray, improved model development, model monitoring visualizations, and new accelerator profiles.

PyTorch introduced ExecuTorch Alpha to enable the deployment of advanced machine learning models, including extensive language models (LLMs), on resource-limited edge devices, such as smartphones and wearables. Leveraging the PyTorch framework, ExecuTorch Alpha provides a complete workflow for deploying AI models on edge devices, from model conversion to optimization and execution. This approach supports the use of efficient model runtimes, the company said, enhancing portability and memory management to bridge powerful AI capabilities with environments constrained by resources. Also, the toolkit maximizes PyTorch’s flexibility and user-friendliness, and focuses on portability and efficient resource management.

Google's artificial intelligence research division, DeepMind, announced a new version of its AlphaFold neural network, which the company said significantly enhances biologists' ability to map the structure of biomolecules. Unlike the first iteration introduced in 2021, which was limited to predicting protein structures, AlphaFold 3 extends its capabilities to include DNA, RNA, ligands, and other biological components, offering greater accuracy than current methods. Powered by an upgraded module known as the Evoformer, and combined with a diffusion network typically used in image generation, AlphaFold 3 enhances the accuracy of biomolecular structure predictions. Google is now offering AlphaFold 3 through a new cloud service, AlphaFold Server, which provides most features free of charge.

German neuromorphic supercomputing company SpiNNcloud Systems announced SpiNNaker2, the first commercially available neuromorphic supercomputer. The supercomputer-level hybrid AI high-performance computer system is based on principles of the human brain. It's a massively-parallel system incorporating 152 ARM processors with powerful native accelerators to support the efficient implementation of sparsity-aware artificial neural networks, symbolic AI, and spiking neural networks. The new SpiNNcloud server board integrates 48 SpiNNaker2 chips, each equipped with a low-power mesh of 152 Arm-based cores. The chips include accelerators, such as distributed graphics processing unit-like units, to enhance computations for neuromorphic, hybrid and mainstream AI models. SpiNNaker2 is the atomic hardware building block of the company's large-scale system, the SpiNNcloud.

Oracle previewed Oracle Code Assist, an AI-driven coding assistant designed to help developers accelerate development and ensure code consistency. The tool, powered by large language models on Oracle Cloud Infrastructure (OCI), is optimized for Java, SQL, and application development. It's designed to provide devs with context-specific suggestions that adhere to an organization's coding standards and practices. And it aims to facilitate the updating, upgrading, and refactoring of code across most modern programming languages. The AI-powered tool promises to transform software development stages from conception to management, the company said, allowing developers to automate routine tasks and receive intelligent coding suggestions. Oracle Code Assist will integrate into the software development life cycle by offering features like code snippets, dependency analysis, error mitigation, test cases, and documentation support, the company said. It will be available as a plugin for JetBrains IntelliJ IDEA or Microsoft Visual Studio Code, Oracle Code Assist will provide tailored feedback for building and optimizing applications.

Meta announced new generative AI-powered features for its Meta Ad Manager, enabling advertisers to automate text and image creation for ads across the company’s platforms, including Facebook, Instagram, Messenger, and third-party websites. These features, part of the campaign management portal's update, will roll out alongside enhancements to Meta's Verified business page verification service, the company said. The updates will be powered by Llama 3, the latest version of Meta's open-source large language model, which has been trained on a dataset more than ten times larger than previously used. The company plans to add new tools and features incrementally, with availability expected globally by year's end. Llama 3 models, with 7 billion and 80 billion parameters, are touted as the most capable in their categories. Additionally, Meta said it will expand its Meta Verified service to offer four subscription tiers, replacing the single-tier system. These tiers will include new tools for profile enhancements and expanded customer support options.

Protect AI Inc., a cybersecurity firm specializing in AI and machine learning (ML), launched Sightline, a dedicated vulnerability database for AI and machine learning. Sightline was designed to offer insights into known and emerging vulnerabilities, enhancing organizations' ability to detect, assess, and remediate AI-related risks promptly, the company said. It also offers penetration testing tools and integrates with Protect AI’s Radar security posture management product to enhance the identification and management of vulnerabilities within a customer's AI supply chain, allowing for swift responses to potential threats. The launch of Sightline adds to Protect AI's portfolio of AI security tools, which includes Guardian, a secure gateway introduced in January to help organizations enforce security policies on machine learning models and prevent malicious code from entering their systems.