CLEVI, a start-up developing artificial intelligence (AI), announced on the 31st that it had developed the generative AI ‘ARI’ using a large language model (LLM). ARI is an omnimodal AI that accepts various types of data, including text, voice, video and images, and produces various forms of output. Omnimodal AI displays answers to questions in various forms, including text, pictures, sound and video, like OpenAI’s ‘GPT4 o’. CLEVI CEO Lee Hwan-ho said, “ARI allows users to ask questions using various materials, including text, photos, voice and video, and also displays a variety of outputs,” adding, “However, due to issues such as server capacity, questions are currently limited to text and voice, and answers to text.”
The company claimed that ARI’s performance surpassed GPT4 based on its own verification results. CEO Lee explained, “ARI applies the chain-of-thought (CoT) technique, which enables it to reason through answers like a human,” adding, “Its performance is better than OpenAI’s GPT4.”
However, it refrained from commenting on its performance compared with the AI ‘DeepSeek R1’, developed by China’s DeepSeek. CLEVI Executive Director Seol Jae-hyeok added, “It is difficult to compare performance directly because much important information about DeepSeek, including specific development costs and parameters, has not yet been disclosed.”
The company also did not disclose the number of parameters or AI semiconductors, which could provide an indication of the size of its AI engine. CEO Lee replied, “We have developed five AIs, but we do not disclose the number of AI semiconductors or parameters because they are company secrets.”
ARI is available in two forms: software (apps) for computers and smartphones. Individuals can use it free of charge, while businesses can use a paid, installable service provided for security purposes.
The company plans to launch ARI’s official service within a month. CEO Lee emphasised, “ARI is currently in trial service,” adding, “We will launch the official service within a month and make it available through various methods beyond the text-based question-and-answer service.”
Choi Yeon-jin IT journalist [email protected]
Source: Hankook Ilbo (Reporter Choi Yeon-jin) | https://www.hankookilbo.com/News/Read/A2025013117210005695?did=NA