Paris-based startup H, founded by former Google employees, received US$220 million in seed round financing without releasing any products. Subsequently, the three founders resigned, which once caused concern in the industry. However, H Company quickly launched its first product, Runner H, an "intelligent agent" artificial intelligence for enterprises and developers, designed to simplify tasks such as quality assurance and process automation. The editor of Downcodes will give you an in-depth understanding of this AI product based on a 2 billion parameter compact large language model and the story behind it.
Paris-based startup H, founded by former Google employees, shocked the industry last summer when it received $220 million in seed round financing without releasing any products. However, shortly after the financing, the company's three founders resigned due to "operational and business differences," raising concerns about its prospects.
However, Company H has not been knocked down by difficulties and recently announced the launch of its first product, Runner H, an "intelligent agent" artificial intelligence for enterprises and developers, designed to handle tasks such as quality assurance and process automation. Runner H is built on the company's proprietary "compact" Large Language Model (LLM) with just 2 billion parameters.
Company H has set up a waiting list for Runner H on its official website. CEO Charles Kantor said that in the coming days, an API will be released to users on the list for them to use H's pre-built agents, and developers can also use it to create their own agents. In addition, using the API will provide access to H-Studio to help users test and manage the operation of these services.
Currently, use of these APIs is free, and a charging model will be introduced later. Despite compact LLMs, building and running AI remains expensive, especially in an increasingly competitive environment. TechCrunch confirmed that Company H is raising Series A financing to support the construction of what Cantor calls "second generation artificial intelligence", in contrast to the "first generation artificial intelligence" represented by companies such as OpenAI.
Cantor said Company H has partnered with several clients in areas such as e-commerce, banking, insurance and outsourcing to refine its products. Runner H will mainly focus on three specific application scenarios: Robotic Process Automation (RPA), Quality Assurance and Business Process Outsourcing (BPO).
RPA is a field that has been around for many years and aims to automate repetitive tasks performed by humans through basic scripting. The goal of Runner H is to be able to perform RPA on modified forms, websites and templates without having to re-write scripts. In terms of quality assurance, Runner H can effectively reduce the maintenance burden of website testing, verify page availability, simulate real user operations, etc.
BPO covers fixes and improvements to improve the billing process and accelerate the ability of agents to obtain and access different data sources. Although the number of parameters has become a competitive focus among basic AI companies, Runner H adopts a different strategy of only 2 billion parameters, emphasizing its cost and operational efficiency.
Company H claims that its compact model outperforms Anthropic's "compute usage" model by 29% on the WebVoyager benchmark, and performs well in model comparisons with Mistral and Meta.
All in all, Company H and its Runner H product, with its unique compact LLM and focused application scenarios, have shown impressive potential in the AI field. Although it faces challenges, its development deserves continued attention. The editor of Downcodes will continue to bring you more technology information.