Generative AI, how will it evolve?

“SLM is gaining popularity due to practical considerations such as computing resources, training time, and specific application requirements,” Thomas said. Over the past few years, SLM has become increasingly relevant, especially in scenarios where sustainability and efficiency are critical. .

According to Gartner, SLM can help most companies achieve task specialization and improve the accuracy, robustness, and reliability of generative AI solutions. Additionally, Gartner said that deployment costs, data privacy, and risk mitigation are key challenges when using generative AI, so SLM offers most companies a cost-effective and energy-efficient alternative to LLM.

Commissioned by startup Hyperscience According to a Harris poll conducted among more than 500 users, According to the study, three out of four IT decision makers (75%) believe that SLM is superior to LLM in terms of speed, cost, accuracy, and ROI.

“Data is the lifeblood of any AI initiative, and the success of a project depends on the quality of the data fed into the model,” said Andrew Jonner, CEO of Hyperscience, which develops AI-based office automation tools. said. Additionally, “surprisingly, three out of five decision makers said they were not utilizing the full potential of generative AI due to a lack of understanding of their own data. “The real potential lies in adopting a custom SLM that can transform document processing and improve operational efficiency.”

Gartner recommends that companies tailor SLM to their specific needs to improve accuracy, robustness, and efficiency. “Improved alignment through task specialization and the inclusion of static enterprise knowledge can reduce costs,” Gartner said. “The hybrid approach is effective and efficient because it can provide dynamic information as needed,” he summarized.

Vendor providing pre-trained generative AI models for the biopharmaceutical industry Y-seop’s CEO Emmanuel Walkener said the future of LLMs in highly regulated industries such as financial services, healthcare and pharmaceuticals is certainly not solid. He also emphasized that using smaller, specialized models can reduce time and energy wasted building large models that are not needed for the current task.

Has potential but is not mature yetis not

“It is unclear how many agents will be able to live up to the high expectations, but we will see an increase in domain-specific AI agents in the coming year,” Gartner’s Chandrasekaran predicted.

According to Forrester Agent AI Architecture is a top emerging technology, but it is still two years away from reaching the high level of automation expected.

Companies want to apply generative AI to complex tasks through AI agents, but this technology mostly focuses on synergy between multiple models, Create search augmentationCustomization through (RAG) is still a difficult technology to develop because it relies on expert knowledge. “Tuning these components for specific outcomes is an unresolved hurdle and prone to developer frustration,” Forrester said in the report.

A recent survey of 4,000 business leaders and technology practitioners across industries According to a survey by Capital One: While 87% believe their data ecosystem is ready to support AI at scale, 70% of technologists spend several hours each day solving data problems.

Nonetheless, the survey results show that business executives are very optimistic about their companies’ AI readiness. In particular, 87% say they have a modern data ecosystem to scale their AI solutions, 84% say they have centralized tools and processes for data management, and 82% are confident in their data strategy for AI adoption. , 78% said they feel prepared to manage the increasing volume and complexity of AI-driven data.

However, by 2025, 75% of companies attempting to build their own AI agents are expected to fail, opting instead for consulting services or pre-integrated agents from existing software vendors. To address the mismatch between AI data readiness and real-world complexity, by 2025, 30% of enterprise CIOs will integrate a chief data officer (CDO) into their IT team to lead AI initiatives, according to Forrester Research. CEOs will recognize that successful AI requires both a solid data foundation and effective stakeholder collaboration, and will rely on CIOs to bridge the gap between technical and business expertise.

Forrester’s 2024 survey also found that 39% of senior data leaders report to the CIO and a similar 37% report to the CEO, and this trend is growing. To drive AI success, CIOs and CEOs must elevate CDOs from mere liaisons to key leaders in AI strategy, change management, and ROI delivery.

In multimodality for interest increase and technology power generation

New use cases for multi-modality, especially images and voice as modalities in both generative AI input and output, will also see greater adoption in 2025.

A subfield of AI Multimodal learning is Improves machine learning through model learning on various data types such as text, images, video, and audio. This approach allows the model to identify patterns and correlations between text and related sensory data.

Multimodal AI expands the capabilities of intelligent systems by integrating multiple data types. These models can process a variety of input types and produce a variety of outputs. For example, GPT-4, which is the basis of ChatGPT, receives both text and image input and outputs text. Open AI’s Sora model is Create an image from text.

Another example is the integration of medical imaging, patient history, and laboratory results to improve diagnosis and treatment. In financial services, multimodal AI can analyze customer phone inquiries and assist contact center staff in resolving issues. Also, in the automobile industry tesla, waymo, Lee Otto Inputs from cameras, GPS, and LiDAR can be integrated with AI to improve autonomous driving, emergency response, and navigation performance for the same company.

Additionally, AI executives are prioritizing business outcomes, organizing their data houses, and AI talent needs to be nurtured I realized that I did. Especially the latter The gap between companies’ generative AI needs and the workforce with the skills to meet them is growing. This is even more important in a growing situation.

“In the coming year, we must face reality to develop an effective AI strategy and action plan,” Forrester said in the report. “Enterprise success in 2025 will depend on strong leadership, strategic improvement, and realignment of enterprise data and AI initiatives to match AI aspirations,” he summarized.

dl-itworldkorea@foundryco.com