This year, artificial intelligence will be dominated by “hybrid” AI and the rise of applications operating from large language models, says the CIO of the investment bank Goldman Sachs Marco Argenti in an interview devoted to outlook for 2024.
But what is hybrid AI?
“Hybrid AI consists of using large models (LLM) as a brain that interprets the prompt. Or as an orchestrator that distributes tasks to a certain number of (AI) models specialized in a specific task,” explains M .Argenti, referring to “basic” models such as ChatGPT.
Marco Argenti, CIO of Goldman Sachs. Goldman Sachs
POCs for 2024
According to Mr. Argenti, it will be too costly for companies other than the world’s richest to implement programs of this scale. Therefore, most companies will simply build smaller neural networks, either in their own data centers or in cloud computing services, trained from their proprietary data.
The idea of specialized models tailored to enterprise data aligns well with the current trend of layering features, such as with LangChain, an open-source framework built on top of generative AI.
“In 2024, most companies will focus on proofs of concept that can provide the best returns,” Mr. Argenti also says about the deployment of hybrid AI.
Coordinating security between different parties is a major problem
In addition to hybrid structures, Argenti sees in 2024 the advent of a new category of third-party applications built from basic artificial intelligence models. “If you think of these basic models as operating systems or platforms, there is a whole world of applications that have not yet emerged around these models,” he says.
“There’s a big opportunity for capital to move to the application layer. I think we’ll see that shift happening most likely as early as next year.”
The need to coordinate security between different parties is a major problem, underlines Mr. Argenti.
“Going forward, it will be important to continue to foster an environment that encourages collaboration between stakeholders, that encourages open sourcing of models, and that develops principles-based rules designed to help manage risks, including including bias, discrimination, safety and soundness, and privacy,” Argenti said.
Source: “ZDNet.com”