Samsung Electronics' institutional sector has confirmed the addition of Google and Microsoft (MS) models to its in-house work assistance artificial intelligence (AI), following Meta. It appears to be an attempt to break away from a closed operation focused on the in-house AI 'DS Assistant,' expanding AI utilization to semiconductor design and development to enhance work efficiency.
On the 9th, according to industry sources, Samsung Electronics' institutional sector recently announced in an internal notice that it would operate its in-house AI in an open multi-model environment. This formalizes the introduction of external open-source AI models in the institutional sector, where work security is crucial. Following the initial introduction of the Meta model last month, the models from Google and MS were added this month. As a result, employees in the institutional sector can now utilize specialized external AI models according to their work characteristics.
The external models that Samsung Electronics has introduced so far for in-house AI include ▲Google 'Gemma3' ▲MS 'Phi-4' ▲Meta 'Llama4.' The latest open AI configurations are structured like a series with various sizes of parameters. Among these, Samsung Electronics primarily adopted a lightweight small language model (sLM) as its in-house AI.
Samsung Electronics has analyzed the unique advantages of each model, allowing for the free selection of AI suited to semiconductor tasks. For dealing with numerical data, Phi-4 is used, while Gemma3 is utilized for analyzing image information. The company introduced the changes in the use of in-house AI to employees, stating that the previously introduced Llama4 has shown remarkable performance in inferring complex knowledge and generating code, noting that "language mixing issues have improved."
Samsung Electronics has previously adopted a 'closed strategy' that primarily relied on its own technology for in-house work assistance AI. The institutional sector permitted the use of ChatGPT at its business sites in March 2023, but this led to incidents where semiconductor information was entered into foreign servers within three weeks. In response, a development of an in-house work assistance AI was initiated at the innovation center (currently the AI center) within the institutional sector, leading to the introduction of the DS Assistant in December that year after resolving security issues.
However, the operation of the DS Assistant in a closed manner has shown limitations in utilizing external data. It cannot be leveraged to enhance competitiveness in semiconductor design, leading to voices within Samsung Electronics saying that "there is a need to embrace external AI and utilize the best tools."
Samsung Electronics operates external AI models in an on-premises (internally constructed) format. This allows the operation of in-house AI using only the data servers established within the business sites, significantly reducing concerns about the external leakage of internal information. A spokesperson for Samsung Electronics noted, "We plan to review and support open-source AI models that aid in improving work efficiency based on various types of work."