AI NewsWaveNC AI Develops Ultra-Light Vision-Language Model VAETKI Vision 3B for Robots
Key points
- NC AI develops ultra-light VLM VAETKI Vision 3B
- About 3 billion parameters, for robots and edge devices
- Game companies continue their push into physical AI
NC AI has developed "VAETKI Vision 3B," an ultra-light vision-language model (VLM) that can be installed directly in robots and industrial equipment. The company announced it on Nov. 3. Techworld News, AI Times, Chosun Biz, Industry News, Digital Today, iNews24, AI Newspaper (인공지능신문), EBN and Digital Daily reported it.
Chosun Biz and Digital Today reported that the model is designed to be installed directly in robots or industrial edge devices with limited computing resources. AI Times and AI Newspaper wrote that the model has about 3 billion parameters. Industry News introduced it as a model installed directly in robots at industrial sites without an external communication network.
Outlets described its performance differently. AI Times wrote in its headline that it outperformed 8B-class models, AI Newspaper said it beat six large global models, and EBN said it outperformed six global models. EBN explained that its distinguishing feature is high performance on major benchmarks of robot and spatial understanding while using fewer computing resources than existing publicly available global models. Only Industry News described the model as the world's highest-performing ultra-light vision AI. Digital Daily said in its headline that a welding-specialized model was developed together with it.
Tech M pointed out that NC and Krafton are expanding into industries beyond games by adding research and development capabilities to the AI technology they built up in game development. Earlier, on the 2nd, Digital Daily reported that Krafton unveiled the robot AI "Ludi 0.2," which was introduced as having stronger ability to recognize human intent and body language.
Glossary A vision-language model (VLM) is an AI model that understands images and text together. Parameters are values an AI model adjusts through training, and are used as a measure of the model's size.