[AI Brief] Team Naver Secures 24 Papers at Global Natural Language Processing Society EMNLP

IT Daily presents 'AI Brief,' which delivers major AI news from Korea and abroad.

In this item, we report global academic recognition of Team Naver's research achievements.

Team Naver said on the 15th that 24 research papers were accepted to EMNLP (Empirical Methods in Natural Language Processing) 2026.

The accepted studies span Korean AI, multimodal, LLM reasoning, and agent technologies, and EMNLP is one of the world's top conferences in the natural language processing field.

The 24 accepted papers involved Team Naver's technical organizations across the board, including Naver, NAVER Cloud, and NAVER LABS Europe, as research participants. The research scope was broad, covering Korean AI, multimodal AI, LLM reasoning analysis, and agent technologies. The multimodal AI category also included speech and video.

Among them, NAVER Cloud focused its research direction on directly defining and solving problems in the process of actual model development and service deployment, rather than improving benchmark scores. In that stream, it produced research findings that identified the cause of degraded instruction-following ability when visual capabilities are added to language models. It also proposed diagnostic metrics for pretraining prediction.

As a regular paper, a benchmark study that redefined the capabilities needed in real-world environments for multimodal AI and evaluated them was also accepted. Team Naver said it conducted research to locate when specific scenes appear in videos lasting several hours. It added that the study showed the limitation of existing AI lies not in the lack of scene recognition, but in the failure to find specific segments within long videos.

Team Naver said it also conducted a benchmark study to evaluate whether omnimodal AI responds with voice tone and speed appropriate to the situation. It further explained that the benchmark showed even models that perform well in text evaluations received failing marks in reality.

Team Naver said it plans to reflect these research achievements in next-generation AI models and services. It also said it aims to further enhance its global AI competitiveness.

The photo is FriendlyAI. FriendlyAI said in an announcement on the 15th that it plans to discuss ways for domestic companies and developers to adopt and use open models together with Ollama.

Jeffrey Morgan and Michael Chang will attend the discussion. Jeffrey Morgan and Michael Chang are co-founders of Ollama.

An event for corporate executives will also be held. The theme of the event is 'Open Models & Sovereignty.'

The event will discuss ways to avoid dependence on specific models and services, as well as ways for companies to secure choice over AI models, data, and infrastructure and secure control over AI models, data, and infrastructure. It will also introduce how to adopt open models based on joint use of Ollama and FriendlyAI, and how to operate open models based on joint use of Ollama and FriendlyAI.

The event is designed to focus on a developer audience and to cover experiences using open models. The agenda was organized around demos.

The event covered how to apply open models to actual products and how to apply open models to AI agents. It also took the form of sharing technical experiences and challenges among developers using open models.

Jeon Byeong-gon, CEO of FriendlyAI, said Ollama and FriendlyAI can be linked by combining their strengths. He also explained that they can be connected from the stage of selecting and using open models through to the stage of stable operation in production environments.

Jeon Byeong-gon, CEO of FriendlyAI, expressed his intention to help more companies review and adopt open models together with Ollama. (Photo: Xennon)

Xennon then said on the 15th that it had launched the Android app for its AI agent integrated portal 'Xena' and announced the transition of 'Xena' to full service. The iOS app for iPhone is scheduled to be released in October.

Xena is an integrated service that allows users to use a variety of AI agents for work and daily life on a single platform. It is designed so users do not need to move separately among multiple AI services for different purposes, and can select and use the agents they need.

Since opening its beta service in last May, Xennon has added new agents and continuously upgraded service features, and with this full launch it has expanded user access by offering mobile apps. Xena currently offers 6 agents: AI Slides, AI Research, AI Data Analysis, AI Document Translation, AI Meeting Notes, and AI Public Policy Guide. Based on functions such as research, document writing and analysis, and meeting summarization, Xena boosts work productivity and has expanded its use into everyday life by including practical services that let users check public policy and support information.

In the full-service stage, it will expand benefits aimed at increasing access to experiences with various AI agents, targeting new subscribers. To that end, new subscribers will receive 10,000 credits every month, which is equivalent to about KRW 10,000 in value. These credits can be used with all currently operating agents.

Source: IT DAILY · Yang Seung-gap
Original: https://www.itdaily.kr/news/articleView.html?idxno=242223

References

• https://www.itdaily.kr/news/articleView.html?idxno=242223