The Week in AI: New Tools from Oracle, Apple, NVIDIA, Snapchat, More

This edition of our weekly roundup of AI products and services includes the PyTorch introduction of ExecuTorch alpha, Oracle's latest Gen AI-enabled customer experience enhancements, Apple's OpenELM advanced language model, HubSpot's relaunched Content Hub, NVIDIA and MIT's new vision language model, OpsVerse's Aiden co-pilot, and new Snapchat AR and ML tools.

PyTorch introduced ExecuTorch alpha, an end-to-end solution for enabling on-device inference capabilities across mobile and edge devices including wearables, embedded devices, and microcontrollers. It's part of the PyTorch Edge ecosystem and enables efficient deployment of PyTorch models to edge devices. Constructed on the PyTorch framework, ExecuTorch Alpha presents a comprehensive workflow for deploying models on edge devices, encompassing model conversion, optimization, and execution. By emphasizing portability and efficient memory management, ExecuTorch Alpha enables the utilization of compact and effective model runtimes across various edge devices. This integration bridges potent AI models with resource-constrained environments.

Oracle unveiled new generative AI-enabled features within its Fusion Cloud Customer Experience application. The new features were designed to deliver more precise responses to customer inquiries, enhance sales force efficiency, and enable marketing departments to evaluate lead quality more effectively in account-based marketing scenarios, the company said. They include Gen AI-assisted answer generation, assisted scheduling for field service, opportunity qualification scoring, and seller engagement recommendations. Oracle said its generative AI services do not share customer data with large language model providers or other customers. Each customer retains sole usage rights for custom models trained on their data, and access is regulated by role-based security controls.

Apple's Machine Learning Research group introduced OpenELM, an advanced open language model. OpenELM utilizes a layer-wise scaling approach to effectively allocate parameters within each layer of the transformer model, resulting in enhanced accuracy. For example, with a parameter allocation of approximately one billion parameters, OpenELM exhibits a 2.36% increase in accuracy compared to OLMo while requiring half the pre-training tokens. Departing from traditional practices, which often only provide model weights and inference code and conduct pre-training on proprietary datasets, Apple's release offers a comprehensive framework for training and assessing the language model on publicly accessible datasets. This encompasses training logs, multiple checkpoints, and pre-training configurations. Fode has been made available to streamline the conversion of models to the MLX library for inference and fine-tuning on Apple devices. This extensive release aims to empower and strengthen the open research community, establishing a foundation for future open research endeavors, the company says.

Software-as-a-Service provider HubSpot relaunched its Content Hub with new GenAI-powered features. The list of new capabilities includes AI Content Creation, encompassing text and image generation; Content Remix functionality, enabling the creation of multiple content variations from a single asset. Content Remix showcased AI's ability to simultaneously generate a blog post, email, ad copy, and social media content derived from a single original asset; Brand Voice feature to ensure consistency in tone across various channels such as social media, blogs, and email; Audio tools for podcast creation and hosting; Post Narration capability to convert text into audio format; Members Blog and Gated Content Library for content management and lead capture.

Pico, a global provider of technology services for the financial markets community, announced the general availability of Corvil Analytics 10.0, and made it accessible for download and deployment via the Pico Client Portal. This latest iteration of Corvil Analytics was designed to harness cutting-edge machine learning (ML) and AI techniques to offer proactive notification and natural language descriptions. These features correlate performance-impacting events, unusual occurrences, and extreme events that influence trading outcomes and infrastructure performance, the company says. With this release, data scientists and quantitative analysts gain access to advanced tools for comprehensive data analysis and operational support. A single Corvil Analytics appliance scales to process up to 7.5 million data points daily, catering to the demanding needs of financial institutions.

Snapchat unveiled a suite of new augmented reality (AR) and machine learning (ML) tools aimed at enhancing brand engagement and advertising experiences on the social platform. Snapchat's AR Extensions enable advertisers to seamlessly integrate AR Lenses and filters into various ad formats within the app, including Dynamic Product Ads, Snap Ads, Collection Ads, Commercials, and Spotlight Ads. Investing heavily in ML and automation, Snapchat aims to streamline the creation of AR try-on assets for brands, facilitating faster and more efficient deployment of interactive experiences. The company has collaborated with industry leaders like Amazon and Tiffany & Co. to enable users to virtually try on products within the app. Now, Snapchat has significantly reduced the time required to develop these AR try-on assets, enabling brands to swiftly convert their 2D product catalogs into immersive try-on experiences.

NVIDIA AI researchers, in collaboration with MIT, unveiled VILA, a vision language model pretrained with interleaved image-text data at scale, enabling video understanding and multi-image understanding capabilities. VILA addresses the need for AI models capable of continuous learning and adaptation in dynamic environments, tackling challenges such as catastrophic forgetting, the company says. VILA distinguishes itself through effective embedding alignment and dynamic neural network architectures, enhancing visual and textual learning capabilities. Pre-trained on large-scale datasets and fine-tuned with Visual Instruction Tuning, VILA has demonstrated significant accuracy gains in visual question-answering tasks, outperforming existing benchmarks and retaining up to 90% of prior knowledge when learning new tasks. This research represents a major stride in advancing Vision Language Models, the company says, promising more effective and adaptable AI systems for various real-world applications. VILA is deployable at the edge by AWQ 4bit quantization and TinyChat framework. 

OpsVerse, a leading provider of DevOps solutions, introduced Aiden, a DevOps copilot driven by generative AI. Aiden was designed to simplify and automate DevOps processes for users. By offering intuitive responses and expert guidance, it enables individuals and organizations to leverage DevOps tools effectively without extensive technical knowledge. Aiden bridges the gap between complex tools and everyday users, empowering them to maximize the potential of their DevOps stacks, the company says. This copilot delivers actionable insights and automated fixes within human-in-the-loop workflows, facilitating swift resolution of production incidents. With Aiden's assistance, developers can enhance key performance indicators, such as Mean Time to Resolution (MTTR) and Mean Time to Detect (MTTD), thereby bolstering customer satisfaction and organizational savings, the company says.

About the Author

John K. Waters is the editor in chief of a number of sites, with a focus on high-end development, AI and future tech. He's been writing about cutting-edge technologies and culture of Silicon Valley for more than two decades, and he's written more than a dozen books. He also co-scripted the documentary film Silicon Valley: A 100 Year Renaissance, which aired on PBS.  He can be reached at