Gauss2 Multimodal AI Reimagines the Relationships Between Language, Images, and Code
The Samsung Developer Conference Korea 2024, a virtual gathering organized by Samsung Electronics today, featured the company’s most recent software advancements and future goals.
This is the eleventh edition of SDC24 Korea, an annual event that Samsung Electronics has hosted since 2014 to interact and work with software developers. Research on software embedded in products including generative AI, software platforms, IoT, healthcare, communications, and data was emphasized at this year’s conference. It also explored the open-source development culture.
The business introduced Samsung Gauss2, the second iteration of its proprietary AI model, during the keynote talk, emphasizing its enhanced performance, efficiency, and range of potential applications.
Samsung Gauss2: A Multimodal Language Model Catering to Diverse Needs
As a multimodal model that integrates language, code, and images, Samsung Gauss2 is the successor to the company’s internal generative AI model, which was presented last year. It promises enhanced performance and efficiency in handling many data kinds at once. Compact, Balanced, and Supreme are the three various models that are offered, each serving a different function.
The Compact model is a little model made especially to function well, even in constrained computing settings. By making the most of the device’s computational capabilities, it provides on-device performance that is optimal. The Balanced approach, on the other hand, aims to provide consistent performance across a variety of jobs by striking a balance between performance, speed, and efficiency. Last but not least, the Supreme model strives for exceptional performance by utilizing Mixture of Experts technology on top of the Balanced model. This enables notable savings in computational expenses during the training and inference phases while preserving high performance and efficiency levels.
Depending on the model, the Samsung Gauss2 supports nine to fourteen languages in addition to several programming languages. To guarantee optimal performance for these supported languages, Samsung has created and integrated its own stabilizing methods for training large language model (LLM) and created a unique tokenizer.
Compared to the top open-source generative AI models currently on the market, the Balanced and Supreme models perform as well as or better in important areas like coding ability and producing responses in several languages, including English and Korean. Furthermore, their hourly processing speed is 1.5 to 3 times faster, allowing for more effective task handling, shorter user wait times, and faster AI response creation.
Customizable for Various Productivity Tasks and Product Development
An internal generative AI model makes customisation simpler and enables the achievement of optimal performance for certain objectives and applications. Employees at Samsung are already using Samsung Gauss for a variety of tasks, taking advantage of its adaptable development capabilities.
The in-house coding helper “code.i” supports the company’s software developers with the help of Samsung Gauss. The Device eXperience (DX) Division’s corporate operations and international research institutes are using the now-upgraded Samsung Gauss2.
Its 60% of all software engineers in the DX Division now use code.i, which has quadrupled in monthly usage since its inception last December. Furthermore, Samsung Gauss Portal is a conversational AI tool that uses Samsung Gauss to help staff members in the DX Division with a variety of office duties, including email authoring, translation, and document summarizing. In April, the service was extended to subsidiaries abroad. Additionally, beginning in August, Samsung will use Samsung Gauss to automatically classify and summarize client calls for call center employees.
By using Samsung Gauss2 to enhance the code’s efficiency, Samsung hopes to steadily boost internal productivity going ahead.service, support multi-modal features including comprehending tables and charts and producing graphics, and improve the Samsung Gauss Portal’s natural language question-and-answer feature.
Furthermore, Samsung will keep extending the reach of its AI-based services across all product lines in order to fulfill its AI vision of “AI for All,” which aims to make consumers’ daily life more convenient and pleasurable. Additionally, Samsung anticipates offering even more improved personalized services through the integration of AI and knowledge graph technology.
Presentations on Topics Including Software Platforms, IoT, Healthcare, Communications and Data
Presentations on the user experience of Samsung’s platforms, including pertinent information on the SmartThings experience and enhancements to several software platforms, were held after the launch of Samsung Gauss2.
29 different technical sessions were subsequently held, which included the following:
- Samsung’s Health Ecosystem Strategy and the Future of Healthcare
- Enhancing the SmartThings Customer VOC Experience with Generative AI
- Using code.i to Improve Your Daily Experience with Samsung Electronics’ AI Coding Assistant TV-based Lifestyle Content Hub
- Using Data and AI Vision Technology to Provide an AI Solution for Samsung Home Appliances
