ARC Solutions, Inc., a deep tech company building the next generation of AI and Web3 resources, recently announced the latest iteration of its high-efficiency AI, Reactor. While the team at ARC may not have been the first to acknowledge environmental sustainability as one of the biggest problems with modern AI technology, the Company’s Reactor AI is evidence that they are among the first to actively address the problem. As such, the team has managed to significantly reduce energy use during AI training while setting new standards for response speed and quality, demonstrating that sustainability and performance in AI are not an either/or challenge.
Humble Beginnings
Founded by TJ Dunham, a 24-year-old entrepreneur from Oklahoma, ARC was initially conceived as a Web3 and blockchain company designed to reduce cryptocurrency trading to three mouse clicks. In the wake of falling victim to a smart contract exploit, the company pivoted toward developing novel methods of securing blockchain infrastructure. By mapping smart contracts into Abstract Syntax Trees (ASTs), ARC’s initial risk detection system could understand code structure and detect potential problems down to the individual node.
A New Approach to Artificial Intelligence
Not satisfied with this solution alone, ARC began implementing an AI component into its system. They set the AI to work on these smart contracts by using Abstract Syntax Development (ASD) to teach the AI how to understand the human intent behind the code it was tasked to analyze. This approach proved both effective and efficient, with ARC safeguarding more than $50 million in value and completing over 1,200 audits. ARC quickly applied what they learned to a large language model, resulting in an AI-first approach and the development of Reactor Mk1.
Reactor Outperforms the Competition
ARC has evaluated the Reactor AI model based on the same industry benchmarks used by industry giants such as Open AI and Meta. In lieu of a firm or even widely agreed upon definition of Artificial General Intelligence (AGI), the industry judges AI model goodness based on performance tests, including MMLU, HumanEval, and BIG Bench Hard (BBH). Without multi-billion dollar investments and football field-sized data centers, Reactor not only manages to keep up with the competition but outperforms many of them. These tests have shown Reactor to be in the same league as the rest of the industry in terms of performance but at a fraction of the cost and a sliver of the environmental impact.
A Sustainable Future for AI
AI technology is infamous for its energy consumption, with OpenAI’s GPT-4 alone requiring more than 50 gigawatt-hours to maintain operations. Reactor requires less than one megawatt hour. By simply running on fewer parameters, ARC’s Reactor model has become a vanguard of sustainable AI, which means it can deliver the benefits of advanced AI without the drastic environmental impact. This level of efficiency is thanks in no small part to innovation in ontological parsing, which allows Reactor to better “understand” human input and intent than other models.
A New Level of Utility for AI Assistance
Backed by its high-performance model, ARC has released Reactor as both a mobile app and a Chrome browser extension to bring AI into the contexts that matter most to users. By bringing Reactor AI directly onto the device and browser level, users benefit from a digital world that is more accessible and more easily navigable with the power of AI, which can now maintain context and continuity in searches, projects, and research work. As such, Reactor is making AI both personal and powerful, which is why the platform is worth a look.
