실행 가능한 객관적인 인사이트를 제공하는 가트너(Gartner)가 최근 2024 생성형 AI 하이프 사이클(2024 Gartner Hype Cycle for Generative AI) 보고서를 통해 향후 5년 내에 조직에 큰 영향을 미칠 잠재력이 있는 기술로 ‘멀티모달(Multimodal) 생성형 AI’와 ‘오픈소스 LLM(대규모 언어 모델)’을 꼽았으며, 향후 10년 내에 주류가 될 것으로 예상되는 가장 잠재력이 높은 기술로 ‘도메인 특화 생성형 AI 모델’과 ‘자율 에이전트’를 꼽았다.
.jpg)
▲Gartner 2024 Generative AI Hype Cycle
Domain-Specific Generative AI Models and Autonomous Agents to Become Mainstream in 10 Years
“Multimodal generative AI and open source LLM will have a major impact on organizations within 5 years”
Gartner, which provides actionable, objective insights, recently announced the latest AI trends to watch in its 2024 Gartner Hype Cycle for Generative AI report.
Gartner cited 'multimodal generative AI' and 'open source LLM (large-scale language model)' as technologies with the potential to have a significant impact on organizations in the next five years, and 'domain-specific generative AI models' and 'autonomous agents' as the technologies with the highest potential to become mainstream in the next 10 years.
“As generative AI evolves to learn across modalities—text, images, audio and video—it has the potential to help capture relationships across diverse data streams, extending the benefits of generative AI across all data types and applications,” said Erick Brethenoux, senior VP analyst at Gartner. “These generative AI models have the potential to “It empowers users to do more without being restricted by the environment,” he said.
“Generative AI will continue to be a challenge for enterprises due to the turbulent and rapid changes in the technology and supply ecosystems,” said Arun Chandrasekaran, senior vice president analyst at Gartner.
He also predicted that “generative AI is currently in the valley of disillusionment, where expectations and interest rapidly decline after the peak of inflated expectations. Once the hype subsides, real benefits will emerge, and capabilities will advance rapidly over the next few years.”
Multimodal generative AI will have a transformative impact on enterprise applications by adding new capabilities and features that were not possible with existing models. This is applicable everywhere AI and humans interact, not limited to specific industries or use cases. Many multimodal models are currently limited to two or three modalities, but will expand to include more modalities in the coming years.
According to Gartner’s report, only 1% of generative AI solutions will be multimodal in 2023, but that number is expected to increase to 40% by 2027. The gradual shift from discrete models to multimodal models will strengthen the interaction between humans and AI, and provide differentiation for products and services that utilize generative AI.
“In the real world, people access and understand information through a combination of modalities, including hearing and seeing,” said Brettnu, Senior VP Analyst. “The reason multimodal generative AI is important is because data is generally multimodal. Combining or combining single modality models to support multimodal generative AI applications often results in delays and inaccurate results, which can lead to a lower quality of experience,” he emphasized.
Open Source LLM is a deep learning-based model that accelerates enterprise value through the adoption of generative AI by democratizing commercial access and allowing developers to optimize models for specific tasks and use cases. It also provides access to developer communities from business, academia, and other research fields to improve models and increase value toward common goals.
“Open source LLMs offer the potential for greater innovation through customization, better control over privacy and security, model transparency, leveraging collaborative development, and reducing vendor lock-in,” said Chandrasekaran, senior VP analyst. “This ultimately gives enterprises smaller, easier-to-train, and less-expensive models to enable business applications and core business processes.”
Domain-specific generative AI models are models optimized for the needs of a specific industry, business function, or task. They improve the set of use cases within an enterprise, providing improved accuracy, security and privacy, and more contextual answers. They can reduce the need for advanced prompt engineering compared to general-purpose models, and reduce the risk of hallucination through goal-oriented training.
“Domain-specific models can accelerate time to value, improve performance, and enhance security for AI projects by providing a more advanced starting point for industry-specific work,” said Chandrasekaran, senior VP analyst. “This will accelerate the broader adoption of generative AI, as organizations will be able to apply these models to use cases where general-purpose models may not perform well enough,” he said.
Autonomous agents are combined systems that achieve set goals without human intervention. They use a variety of AI technologies to perform a series of tasks, such as identifying environmental patterns, making decisions, executing actions, and generating results. They also have the potential to learn from their environment and improve over time, allowing them to handle complex tasks.
“Autonomous agents represent a major shift in AI capabilities,” said Brettnu, Senior VP Analyst. “They can improve business operations, enhance customer experiences, and enable new products and services through independent operational and decision-making capabilities. They drive competitive advantage by reducing costs, and help organizations shift their workforce from performing simple tasks to overseeing them.”
Gartner clients can learn more in the 2024 Generative AI Hype Cycle. They can also learn more in the free Gartner webinar, What Makes Mature Organizations Different for AI Success.