Nvidia's research indicates that Small Language Models (SLM) could become pivotal for the future of the artificial intelligence sector, despite ongoing trends in investments towards Large Language Models (LLM).
Differences between SLM and LLM
Small Language Models (SLMs) are trained on up to 40 billion parameters and excel at a narrow range of tasks while consuming significantly less resources. In contrast, Large Language Models (LLMs) incur substantial infrastructure costs. For instance, OpenAI's CEO Sam Altman remarked that using ChatGPT costs the company tens of millions of dollars.
Investments and Economic Impacts
Despite the high costs associated with LLMs, companies such as OpenAI are actively attracting investments. However, if investment levels in LLMs remain unchanged, this could negatively impact the U.S. economy. The AI sector raised $109 billion in 2024, and this year $400 billion has already been spent on infrastructure. If more data centers are not built, it may lead to declining investor interest.
Recommendations for AI Optimization
Nvidia researchers suggest utilizing SLMs to enhance the specialization and efficiency of AI agents. They recommend creating modular systems where LLMs are used only for complex tasks, allowing for resource savings and increased competitiveness.
In the face of increasing instability in the artificial intelligence sector, Small Language Models represent a more economical and efficient alternative to Large Language Models. Continued research and development in this area is anticipated to help avert potential crises within the industry.
