Orby Reveals ActIO: A Leap Forward in Agentic AI with Groundbreaking Performance
In a significant stride within the AI domain, Orby AI, a pioneering force in generative AI technologies for enterprise, has announced the launch of ActIO. This introduction marks the debut of the most proficient large action model (LAM) AI foundation to date, achieving unprecedented results on the Large Action Model Benchmark.
In collaboration with the Ohio State University’s Natural Language Processing (NLP) group, Orby has made substantial progress in advanced AI techniques, notably in visual grounding. This advancement empowers AI agents to correlate visual inputs with linguistic understanding, a capability encapsulated in the newly developed technique known as UGround. Their joint research paper, “Navigating the Digital World as Humans Do: Universal Visual Grounding for GUI Agents,” delves into this innovation, now integral to Orby’s ActIO foundation LAM.
Traditional large language models have faced challenges in bridging the divide between visual data and textual understanding, often overlooking intricate details or entirely misinterpreting information. Through its partnership with OSU, Orby has enhanced machine capabilities, enabling them to recognize and comprehend the significance of visible elements within specific contexts and tasks.
“The advancements we’re witnessing and contributing to within the AI realm are monumental, surpassing the transition periods to mobile and even the web,” expressed Will Lu, Co-Founder and CTO of Orby. “The next generation AI systems require an adeptness in processing and interpreting visual content, understanding objects, scenes, relationships, and language. We’ve achieved just that,” Lu added.
Yu Su, Assistant Professor in the Department of Computer Science and Engineering at Ohio State University, echoed the sentiment, emphasizing the vast potential that this milestone with Orby unlocks.
In a move towards open innovation, Orby and OSU have made the new visual grounding model accessible on HuggingFace, enabling developers to incorporate this model into various applications.
Introducing ActIO
Setting a new precedent, ActIO emerges as the first commercially available and patented LAM AI foundation, distinct in its advanced capacities for decision-making, planning, and responding to dynamic alterations in scenarios.
Boasting the highest accuracy and success rates among AI agents, ActIO excels in analyzing complex situations to make informed decisions. ActIO astutely automates intricate and repetitive enterprise workflows with minimal oversight, serving as a multimodal generative AI foundation model crafted for the intricacies of enterprise use cases.
In comparative benchmarks, including the VisualWebBench, ActIO has surpassed leading models like GPT-4o, Gemini 1.5 pro, and Llava 1.6-34B, showcasing superior performance in visual web understanding.
ActIO’s proficiency extends to supporting GUI agents, where it leads as the top-performing system. In extensive tests, ActIO convincingly outperformed GPT-4, showcasing a 25% improvement in accuracy across prominent digital agent benchmarks. This pivotal enhancement signifies Orby’s AI agent’s capacity to adeptly adjust to website UI changes, dynamic content, and unstructured data sets, allowing users to modify automation steps through natural language seamlessly.
About ORBY AI
Based in Mountain View, California, Orby AI stands at the forefront of revolutionizing enterprise operations. Leveraging proprietary generative AI, Orby AI’s platform enables the automation of complex processes at scale, bypassing the limitations of traditional systems. With roots in Google and UiPath, Orby’s team of AI veterans has secured $35 million in investments from leading entities such as New Enterprise Associates (NEA), Pear Venture Capital, WndrCo, and Wing VC.
For further details on Orby’s trajectory altering ActIO model and their vision for the future of enterprise automation, contact Brad Day at brad@orby.ai or call +1 (408)-505-5399.
Market News and Data brought to you by Benzinga APIs.
