Only the richest companies like OpenAI can build "base models" like GPT-4. The rest of the business world will need to build smaller, more focused programs that tap into these base models.

Marco Argenti, Chief Investment Officer at investment bank Goldman Sachs, stated in an annual outlook interview that this year will be dominated by "hybrid" AI, as well as the rise of applications running on top of large language models.

"Hybrid AI refers to using these large models as a brain to interpret prompts and user needs, or as orchestrators to assign tasks to specialized worker models for specific tasks," Argenti said when referring to "base" models like ChatGPT.

He believes that building such massive programs is too expensive for most companies, except for the wealthiest ones. Therefore, most businesses will settle for building smaller neural networks trained on their proprietary data in their own data centers or cloud computing services.

The concept of fine-tuning professional models with enterprise data aligns well with the current trend of connecting functionalities, such as through open-source frameworks like LangChain, which is built on generative AI.

When discussing the potential launch of hybrid AI, Argenti said, "By 2024, most companies will focus on concept validation that shows the highest return."

In addition to hybrid structures, Argenti also sees a new class of third-party applications emerging on top of base models in 2024.

"If you think of these [base models] as operating systems or platforms, there is a whole world of applications around these models that hasn't truly emerged yet," Argenti said.

"The shift of capital to the application layer and toolset layer is a great opportunity. I think we will see this shift happen, possibly as early as next year."

Argenti emphasized that coordinating security among different parties is a major concern.

"Looking ahead, continuing to foster an environment that encourages collaboration among players, opening up model source code when appropriate, and establishing principled rules to help manage potential risks, including bias, discrimination, safety and robustness, and privacy, will be important," Argenti said. "This will allow technology to progress so that the United States can maintain its leadership position in the development of artificial intelligence."