Naver unveiled three lightweight models as an open source and predicted the launch of the reasoning model in the primary half. Through this, it should begin in earnest the ‘On Service AI’ strategy that applies AI to Naver’s service. It also emphasized that it’s a true ‘Sovereign AI’ by developing self -development, not in partnership with overseas big tech.
Naver Cloud (CEO Kim Yoo -won) held a gathering at Yeoksam Square in Seoul on the twenty third and announced the event of artificial intelligence (AI) and the on -service AI strategy.
Kim Yu -won, CEO of Naver Cloud, said, “Naver is exclusive in that it is predicated on large quantities of service and contents in addition to the technology required for Sovereign AI.”
On -service AI is an idea introduced on the ’24’ event last 12 months, which refers to melting AI into Naver’s all services. The event of a light-weight model explained that it was essentially the most efficient judgment to mount the AI function for every service beyond simply reducing the dimensions.
The name of the lightweight model is ‘hyperclobar X seed’, and the parameters consist of 3B, 1.5B, and 0.5b.
First, the biggest 3B model was introduced as a multimodal model specialized in understanding image related to Korean and domestic culture. It has the power to process and analyze various visual data.
The 1.5B model explained that it is beneficial for translating between major languages equivalent to English and Japanese. It is usually useful for business or chatbot services, which might be talked with a particular tone or tone. Within the Korean specialized benchmarks equivalent to ‘KMMLU’ and ‘Best’, it is healthier than the massive tech model equivalent to Alibaba’s ‘Q1’ and Google’s ‘Gemma’.
Finally, the micro 0.5b model was optimized for Ondivis AI. You’ll be able to run locally without a web connection in your phone, smart home, and wearable devices. He added that it could be useful for startups because it has a low price and infrastructure burden.
Hyperloba X seed model group will probably be released on the twenty fourth through an open source. Industrial use can be possible. Kim said, “It’s a call that’s more like a public interest to revitalize the national AI ecosystem somewhat than business purposes.”
The reasoning model will probably be released in the primary half. It also said that because it strengthened the power to think itself based on the small parameters, it could help to introduce domestic AI agents. Nonetheless, it just isn’t decided whether to be released as an open source.
The reasoning model is put into Naver service and acts as a form of agent. Based on Antropic’s ‘Model Context Protocol (MCP)’, which has recently grow to be a global standard for agent construction, the corporate plans to attach external data or tools to act as an agent.
For instance, should you ask the Clova X chatbot to “organize the precise contents and record it within the ‘Fall’ app,” you’ll have the opportunity to connect with the Kneness and perform a command. Seong Nak -ho, general manager of Naver Cloud Technology, said, “With MCP alone, it could actually be a powerpoint in addition to notification, and even connected to cars and fridges.”
Meanwhile, a big -scale update of the flagship model ‘Hyperloba X’ will probably be carried out soon.
It is supplied with the ‘audio function’ mentioned since last 12 months. In comparison with other AI chatbots, it’s characterised by ‘natural tone’, and it should be accomplished in the primary half of the 12 months and will probably be applied to the product within the second half.
It also adds video understanding. This lets you seek for specific sections in natural language within the video. The launch point was not set.
Recently, loads of models have been poured out in the worldwide market, and the popularity of Hyperclova X has also been significantly reduced.
Due to this fact, the technology report will probably be released through the Hyperclopa X Update. Kim explained, “There may be a giant update to make a leap as a representative model.”
Sung Nak -ho also explained that Naver’s AI capability was not falling. He said, “The technology and talent that might be comparable to big tech, equivalent to multimodal models and image creation, are all equipped,” he said.
At the top, it was re -emphasized by Naver’s slogan Sovereign AI. Kim Yoo -won said, “If you happen to only put a trademark on foreign technology, it should not be a real Sovereign AI.” “Naver’s AI will show the ability to vary the actual service.”
This appears to be geared toward other domestic large firms which are developing models and services through partnerships with open AI, Microsoft, and Antrofac.
By Jang Se -min, reporter semim99@aitimes.com
