[AI Revolution](119) Coxwave: "Conversations with Chatbots Are Key to Improving AI Performance"

Analysis of Human and AI Conversation Data
Proposing AI Chatbot Improvement Strategies
Expanding Beyond Text to Multimodal Analysis
"Analyzing AI-to-AI Interactions to Unveil the Black Box"

With the advent of generative artificial intelligence (AI) technology enabling natural language conversations between humans and machines, the interaction data exchanged between them is gaining attention. It is emerging as a key factor in improving AI models.

Generative AI company Coxwave has focused on this change and developed the large language model (LLM) analysis platform ‘Align AI.’ Their ambition is to serve as the ‘helmsman (Cox)’ that helps find the direction for AI chatbots.

Coxwave conceived Align AI out of a desire to improve based on experiences operating AI services. While running the image editing service ‘Hama’ and image search service ‘Enterpix,’ they found it difficult to understand how users were using these services and what improvements they wanted.

Kim Ki-jung, CEO of Coxwave, is giving an interview to Asia Economy at the Yeouido office building in Seoul on the 26th. Photo by Hyunmin Kim kimhyun81@

Kim Ki-jung, CEO of the company, thought, "Just as Google Analytics analyzes user dwell time and purchase frequency on web pages, an analysis tool suited for AI is necessary." After selling Hama and Enterpix to the font company Sandoll in 2022, Coxwave developed Align AI.

Align AI focuses on conversations users have with AI. These conversations reveal various information about users, including service satisfaction. Unlike the existing Chopin app, with ChatGPT, users exchange dialogues such as, "I don’t like the clothes you just recommended; find me checkered patterns instead of stripes." Through conversation analysis, Align AI suggests improvements to AI service developers and operators. It monitors whether inappropriate or unexpected responses occur, whether retrained data is well reflected, and recommends enhancing AI chatbots with frequently used features. Clients of Coxwave include Danal, Samsung Fire & Marine Insurance, and Meritz Fire & Marine Insurance, and they are about to sign a contract with one of the Big Four accounting firms.

As the generative AI service market grows, it presents an opportunity for Coxwave. Approximately 20-25% of the software (SW) market size is in quality assurance and quality control (QA/QC) fields that analyze and test software. Applying this to the generative AI market, it is estimated that a global market worth about 40 trillion won will open by 2030.

To secure market leadership, CEO Kim is also targeting overseas markets. Last month, they formed a strategic partnership with India’s Fundamento, which develops AI contact centers (AICC). They plan to integrate Align AI into Fundamento’s AICC to improve the productivity of customer service agents. Based on this, they will pursue entry into the Indian and global markets.

Technically, they plan to expand to multimodal model analysis. This means analyzing not only text exchanged between AI and humans but also image, video, and voice data. Furthermore, they aim to analyze interactions between AIs themselves. AI often interacts with other AIs by asking questions before providing answers. Kim believes that analyzing this can contribute to solving the long-standing ‘black box’ problem of AI. AI is called a black box because the process it goes through to produce results is not clearly known.

CEO Kim emphasized, "The emergence of generative AI represents a paradigm shift in the interface through which humans and computers communicate," adding, "Now is the right time to seize the opportunity of this change."