Last Week in AI: Hyperscience, Cohere, Zencoder, Genmo, Asana, and More -- Pure AI

Last Week in AI: Hyperscience, Cohere, Zencoder, Genmo, Asana, and More

By Pure AI Editors
10/28/2024

This edition of our roundup of AI products and services announced last week that didn't make splashy headlines, but should be on your radar, includes Zencoder's AI Agents, Hyperscience's Hypercell platform, Cohere's Aya Expanse, Genmo's Mochi 1, Asana's AI Studio, and more.

IBM Corp. launched the Guardium Data Security Center, a comprehensive platform designed to address data security challenges posed by hybrid cloud, artificial intelligence, and quantum computing. The platform provides a unified view of an organization’s data assets, enabling security teams to manage data monitoring, AI security posture, cryptography, and compliance from a single dashboard. Guardium Data Security Center includes Guardium AI Security, aimed at safeguarding AI deployments from vulnerabilities and unsanctioned “shadow AI” risks. It also features Guardium Quantum Safe, a tool developed with IBM Research to protect encrypted data from future quantum threats, leveraging IBM’s post-quantum cryptography algorithms.

Hyperscience Inc. launched an updated version of its AI-powered Hypercell platform, aimed at accelerating automation for back-office document processing and complex workflows. The update enhances Hypercell’s machine learning and artificial intelligence capabilities, with deep learning models for efficient long-form document extraction and improvements in model lifecycle management. Hyperscience says Hypercell’s AI-first approach addresses the limitations of traditional, rule-based automation technologies, which often fail to adapt to diverse document types and complex processes. The platform now offers automation rates of 98% and accuracy of 99.5%, alongside enhanced governance features to ensure compliance and traceability, enabling organizations to confidently scale digital transformation initiatives.

Cohere for AI, the nonprofit research lab of Cohere Inc., unveiled Aya Expanse, a new family of high-performance multilingual AI models designed to enhance language model research across diverse languages. Aya Expanse includes models with 8 billion and 32 billion parameters, covering 23 languages such as English, Arabic, and Chinese, and is available with open weights on Kaggle and Hugging Face. Aya Expanse incorporates some innovative techniques, including synthetic data, human feedback, and model merging, to address gaps in language representation and avoid common issues in multilingual AI. The model outperformed major rivals, including Google’s Gemma 2 and Meta’s Llama-3.1, in multilingual performance tests.

AI startup Zencoder launched a new suite of generative AI coding agents, aimed at boosting developer productivity through intelligent code generation, self-repair, testing, optimization, and documentation. Zencoder’s AI agents are designed to integrate into development workflows, providing multistep coding capabilities beyond traditional AI assistants. Zencoder’s standout feature, “Repo Grokking,” enables AI agents to deeply analyze a company’s codebase, improving contextually relevant code suggestions and automating bug fixes. Agents self-diagnose errors using syntax analysis, refining code to minimize AI generation issues. Supporting Java, JavaScript, Python, and other languages, Zencoder also integrates with Visual Studio Code and JetBrains.

AI startup Haiper Ltd. unveiled Haiper 2.0, its latest video generation model capable of creating ultra-realistic video clips based on user prompts. The new model, an upgrade from Haiper’s previous release, promises faster generation speeds and a more lifelike style, with an upcoming update to boost video resolution to 4K quality. Built on the DiT architecture, Haiper 2.0 combines diffusion and transformer AI technologies to deliver enhanced performance and hardware efficiency. With Haiper 2.0, users can generate six-second clips from prompts or reference images, and extend existing clips by two seconds. The model enters a competitive market with rivals such as Adobe and OpenAI also introducing advanced video editing and generation tools.

AI content generation platform Genmo Inc. unveiled Mochi 1, an open-source model for text-to-video generation. Genmo describes Mochi 1 as a step toward building the "right brain" of artificial general intelligence, bringing creativity to the forefront of AI capabilities. With advanced motion quality and strict adherence to user instructions, Mochi 1 aims to reduce "daydreaming" in AI models by following precise text prompts. Users can try Mochi 1 in Genmo’s new hosted playground, and model weights are available on Hugging Face. Built with a 10-billion-parameter Asymmetric Diffusion Transformer (AsymmDiT), Mochi 1 generates realistic videos up to 5.4 seconds long at 30 frames per second, initially at 480p with a 720p HD version slated by year’s end.

UiPath Inc. announced a preview of its upcoming Agent Builder tool at its annual user conference, UiPath Forward 2024, setting the stage to blend generative AI with its robotic process automation (RPA) technology to advance enterprise automation. Launching in preview this December, Agent Builder will enable developers to design AI-powered agents that can collaborate with traditional RPA bots, aiming to automate complex tasks, the company says. The new approach, which the company calls " agentic automation," allows AI agents to make intelligent decisions and carry out multistep processes, expanding beyond traditional, rule-based automation. According to UiPath CEO Daniel Dines, this evolution will empower businesses to automate end-to-end processes, with strict governance controls managed through UiPath’s platform.

Asana Inc. introduced Asana AI Studio, a no-code tool designed to boost productivity through customizable AI-driven workflows. Available in early access, AI Studio enables companies to create "smart workflows" that integrate seamlessly within Asana’s work management platform, facilitating task automation and collaboration. AI Studio leverages the platform’s Work Graph technology to understand team structures, task dependencies, and organizational priorities. Unlike typical AI tools, the company says, AI Studio operates like a virtual teammate, assisting with routine tasks across departments, including marketing, IT, and operations.

Freshworks Inc. announced the release of Freddy AI Agent, an autonomous, no-code AI tool designed to enhance customer and employee support. Deployable in minutes without consulting fees, Freddy AI Agent was designed to autonomously resolve many service requests by leveraging existing resources such as FAQs and company websites. Integrating with such Freshworks platforms as Freshdesk and Freshservice, Freddy AI provides 24/7 human-like support across multiple languages, freeing staff to handle complex issues and increasing productivity. Early adopters, including Porsche eBike and Live Oak Bank, report significant efficiency gains from the new tool.

Generative AI startup Runway AI Inc. launched Act-One, a new feature that allows users to replicate their own facial expressions onto AI-generated video characters using just a smartphone camera. Available to Runway account holders with credits, Act-One aims to make advanced facial animation more accessible to creators. Act-One works with the Runway's Gen-3 Alpha model, which supports text-to-video, image-to-video, and video-to-video generation. By using Act-One, animators, filmmakers, and game developers can transpose their recorded facial expressions to AI characters, enabling more cinema-like realism and nuanced storytelling. Runway emphasized that Act-One includes safeguards to prevent misuse, such as restrictions on unauthorized content featuring public figures and tools for verifying voice usage rights.