Last Week in AI: ZenaDrone, TestSprite, SuperNova-Medius, Palmyra X 004, More -- Pure AI

Last Week in AI: ZenaDrone, TestSprite, SuperNova-Medius, Palmyra X 004, More

By Pure AI Editors
10/14/2024

This edition of our roundup of AI products and services announced last week that didn't make headlines, but should be on your radar includes ZenaTech's ZenaDrone, TestSprite's automated test platform, Arcee AI's SuperNova-Medius, new chips from AMD, and more.

AI drone solutions provider ZenaTech launched ZenaDrone IQ Nano, the latest addition to its IQ series of drones. Designed for warehousing and logistics sectors, the IQ Nano was developed to enhance productivity and reduce costs by automating indoor inspections, inventory management, and security monitoring, the company said. The compact 10x10-inch drone is equipped with advanced sensors, high-quality cameras, and obstacle avoidance technology. The drone performs tasks, such as bar code scanning, facility maintenance, and 3D mapping. It boasts 20-minute flight time and comes with an automatic charging pad. The drone also comes with indoor GPS technology meant to enhance safety and precision in confined spaces. The IQ Nano is built for durability with a carbon-fiber shell and propeller guards, making it suitable for industrial environments, the company said.

AMD announced updated versions of its chip portfolio in an effort to position itself as a key provider of data center technology for organizations utilizing "Big AI." In a bid to challenge Nvidia’s dominance, AMD introduced new Instinct MI325X GPUs and 5th Gen EPYC CPUs, along with AI networking solutions. The Instinct MI325X, built on AMD’s CDNA 3 architecture, boasts industry-leading memory and bandwidth, delivering up to 1.4 times the inference performance of Nvidia’s H200. AMD’s 5th Gen EPYC processors, codenamed "Turin," feature up to 192 cores and are designed to handle demanding AI workloads, offering significant performance boosts over competitors. AMD’s Pensando DPU and AI NIC solutions also target hyperscalers, while the updated ROCm software stack enhances AI training and inference performance. AMD expects the data center AI accelerator market to grow to $500 billion by 2028.

TestSprite, provider of an AI-powered software testing platform, announced early access for developers to a fully automated solution for back-end and front-end quality assurance. The company's namesake platform automates test case generation by analyzing software context, documentation, and custom instructions, minimizing manual input from users. The platform drafts natural-language test plans for review before generating, scheduling, and executing test scripts. The platform integrates with tools such as GitHub and supports continuous development workflows. TestSprite will also expand to test large language models and AI agents, with a broader release planned for late October. Developers can join the early access waitlist now through the company’s website.

Enterprise-focused AI startup Writer Inc. introduced Palmyra X 004, a proprietary large language model (LLM) designed to power AI applications and autonomous agents. The model was developed using synthetic data, which allowed Writer to cut production costs while maintaining high performance, the company said. Palmyra X 004 features automatic data integration and retrieval-augmented generation (RAG), enabling it to enhance training with real-time data. With a 128,000-token context window, it matches the capacity of OpenAI’s GPT-4o. Palmyra X 004 supports external tool calling, allowing AI agents to automate complex workflows with minimal human intervention. Writer says this agentic AI capability positions the model to help enterprises streamline internal and customer-facing tasks. The model has already achieved top-10 results in Stanford University’s HELM evaluations, scoring 86.1% on HELM Lite and 81.3% on HELM MMLU. Notable customers using Writer’s technology include Intuit, Uber, L’Oreal, and Accenture.

Arcee AI unveiled SuperNova-Medius, a 14-billion parameter small language model (SLM) that aims to match the performance of larger models while minimizing computational costs. This release follows the company's previous models, SuperNova-70B and SuperNova-Lite, showcasing Arcee’s push to balance size and efficiency in AI design. SuperNova-Medius leverages advanced optimization techniques, including logit distillation from Llama 3.1 405B and cross-architecture adaptation using Qwen2.5-14B. Its design allows it to excel in instruction-following (IFEval) and complex reasoning (BBH) tasks, outperforming similarly sized models. "We are challenging the traditional notion that larger models are inherently better," the company said in a statement, emphasizing that SuperNova-Medius offers high-quality results without requiring extensive computational resources. With support for diverse domains and multiple languages, SuperNova-Medius is positioned for real-world applications, from startups to educational institutions. Its efficiency and scalability make it a cost-effective solution for organizations seeking reliable generative AI tools.

Atlassian announced the general availability of Rovo, its new generative AI assistant designed to enhance enterprise knowledge discovery and workflow automation. Initially introduced in May with a closed beta, Rovo is powered by Atlassian Intelligence and integrates with tools such as Google Drive, SharePoint, Figma, and GitHub. Rovo provides AI-powered search, a context-aware chatbot, and "Rovo Agents" that act as virtual teammates to streamline tasks. Rovo Agents can assist with complex tasks, from writing bug reports and release notes to generating code for developer pull requests. Teams can also create custom agents to extend Rovo’s functionality. According to Jamil Valliani, Atlassian’s head of product for AI, more than 20 agents will be available at launch, with more planned through marketplace partnerships. Atlassian aims to make Rovo accessible from web browsers, allowing users to query internal company data directly from everyday tools like Google Docs, enabling faster and smarter decision-making across organizations.

Researchers from University College London, the University of Liverpool, Shanghai Jiao Tong University, The Hong Kong University of Science and Technology (Guangzhou), and Westlake University unveiled OpenR, an open-source framework designed to enhance the reasoning capabilities of large language models (LLMs). Inspired by OpenAI’s o1 model, OpenR integrates reinforcement learning, test-time computation, and process supervision to advance LLM performance. It leverages tools such as policy learning, inference-time-guided search, and Markov Decision Processes to improve reasoning through step-by-step optimization. OpenR uses Process Reward Models (PRMs) to provide detailed feedback throughout the reasoning process, which helps the model fine-tune decision-making beyond relying solely on final outcomes. In experiments using the MATH dataset, the framework achieved a 10% improvement in reasoning accuracy compared to traditional methods. Key techniques include multi-search strategies like "Best-of-N" and beam search, which explore multiple reasoning paths during inference, outperforming simpler methods such as majority voting. OpenR’s reinforcement learning capabilities also demonstrated effective performance in online policy learning scenarios. OpenR aims to promote community collaboration with its open-source platform, encouraging further development in AI reasoning. Researchers plan to expand its capabilities to cover more complex reasoning tasks and optimize its inference strategies in future releases.

The team behind Slingshot, a work management platform by Infragistics, introduced new AI-powered tools aimed at helping teams quickly derive actionable insights from company data. The platform was designed to centralize data from various departments, platforms, and channels, offering users automated performance insights in seconds. The AI capabilities allow teams to ask business-related questions and receive rapid, data-driven answers, boosting productivity by eliminating manual data analysis, the company said. Slingshot AI also generates visualizations to aid in performance tracking and trend analysis, empowering everyday business users to make informed decisions. Slingshot is available on desktop and mobile platforms.