Gauss2 Multimodal AI: Code, Images, Language

 

Rethinking the Connections Between Language, Images, and Code with Gauss2 Multimodal AI

Today, Samsung Electronics hosted a virtual event called the Samsung Developer Conference Korea 2024, where the firm showcased its latest software developments and future objectives.

Samsung Electronics has been hosting SDC24 Korea, an annual event since 2014 to engage and collaborate with software developers, for the eleventh time. This year's conference placed a strong emphasis on research on software embedded in goods, including generative AI, software platforms, IoT, healthcare, communications, and data. The culture of open-source development was also examined.

During the keynote address, the company unveiled Samsung Gauss2, the second version of its proprietary AI model, highlighting its improved performance, effectiveness, and possible variety of uses.

A Multimodal Language Model Meeting Various Needs: Samsung Gauss2

Samsung Gauss2, a multimodal model that combines language, code, and visuals, is the replacement for the company's in-house generative AI model that was unveiled the previous year. It offers improved efficiency and performance while managing multiple types of data simultaneously. Three different types are available, each with a distinct purpose: Compact, Balanced, and Supreme.

The Compact model is a small model designed specifically to perform well even in computer environments with limited resources. It offers excellent on-device performance by maximizing the device's computing capacity. By balancing performance, speed, and efficiency, the balanced approach, on the other hand, seeks to deliver consistent performance across a range of jobs. Finally, by building upon the Balanced model using Mixture of Experts technology, the Supreme model aims for outstanding performance. This maintains high performance and efficiency levels while allowing for significant computational cost savings during the training and inference stages.

In addition to a number of programming languages, the Samsung Gauss2 supports nine to fourteen languages, depending on the model. Samsung has developed a unique tokenizer and integrated its own stabilizing techniques for large language model (LLM) training in order to ensure optimal performance for these supported languages.

The Balanced and Supreme models outperform the leading open-source generative AI models available on the market in key categories such as coding proficiency and multilingual response generation, including English and Korean. Additionally, their hourly processing speed is 1.5 to 3 times faster, which enables faster AI answer creation, lower user wait times, and more efficient task handling.

Adaptable to Different Product Development and Productivity Tasks

For specific goals and applications, an internal generative AI model facilitates customization and allows for the attainment of optimal performance. Samsung Gauss's flexible development capabilities are already being utilized by Samsung employees for a range of tasks.

With the aid of Samsung Gauss, the company's software engineers are supported by the in-house coding assistant "code.i." The updated Samsung Gauss2 is being used by the corporate operations and foreign research institutes of the Device eXperience (DX) Division.

Since its launch in December of last year, code.i has quadrupled in monthly usage, with 60% of all software engineers in the DX Division currently using it. Additionally, Samsung Gauss Portal is a conversational AI application that leverages Samsung Gauss to assist DX Division employees with a range of office tasks, such as document summarization, translation, and email composition. The program was expanded to overseas subsidiaries in April. Furthermore, starting in August, Samsung will deploy Samsung Gauss to automatically categorize and compile customer calls for call center staff.

Samsung intends to gradually increase internal productivity in the future by leveraging Samsung Gauss2 to improve the code's efficiency.service, enhance the Samsung Gauss Portal's natural language question-and-answer function, and offer multi-modal features including understanding tables and charts and creating graphics.

Furthermore, in order to realize its AI goal of "AI for All," which seeks to improve the convenience and enjoyment of consumers' daily lives, Samsung will continue to expand the reach of its AI-based services across all product lines. Furthermore, Samsung expects to provide even more enhanced personalized services by combining knowledge graph technology and artificial intelligence.

Presentations on Data, Communications, IoT, Software Platforms, and Healthcare

Following the release of the Samsung Gauss2, presentations were given on the user experience of Samsung's platforms, including relevant details on the SmartThings experience and improvements to a number of software platforms.
  • Thereafter, 29 distinct technical sessions were conducted, encompassing the following:
  • The Future of Healthcare and Samsung's Health Ecosystem Strategy
  • Using Generative AI to Improve the SmartThings Customer VOC Experience
  • Samsung Electronics' AI Coding Assistant TV-based Lifestyle Content Hub: Enhancing Your Everyday Experience with Code.i Data and AI Vision Technology to Offer an AI Solution for Samsung Home Appliances

Post a Comment

0 Comments