In the rapidly evolving world of technology, GenAI, or generative AI, has emerged as one of the biggest surprises of 2023. What was once a topic confined to the realm of IT departments has now made its way into the boardrooms of businesses across various industries. The AI revolution is still in its early stages, but surveys consistently show that GenAI is set to dominate business priorities in the coming year.
As we look ahead, it becomes clear that AI will play an increasingly vital role in enterprise digital transformation. It will serve as a critical driver of innovation rather than an afterthought. At Orion Innovation, we have spent considerable time incubating new ideas and co-innovating with clients on their 2024 GenAI strategies. Based on our experiences and observations, we offer six predictions for enterprise-level GenAI adoption in the near future.
The year 2023 was primarily focused on GenAI experimentation and proof of concepts (POCs). Businesses explored various use cases, ranging from internal productivity enhancements to incorporating GenAI-enabled features into existing platforms. During this period, enterprises were still in the process of grasping the basics of the technology – its potential and limitations. As we move into 2024, we will witness an increase in POCs, but many use cases will advance to the pilot launch stage.
Most GenAI projects today follow the RAG (retrieval augmented generation) architecture pattern, leveraging cloud-based solutions. Major cloud providers offer the necessary basic tech stack, making it relatively easier to build RAG-patterned AI POCs. However, off-the-shelf projects often lack accuracy and developers face challenges such as document chunking, semantic caching, security concerns, audit trails, cost reporting, observability, and more.
In the near future, we expect cloud providers to simplify the adoption of RAG-driven GenAI services. They will add automated tools and resources to overcome the current deployment challenges, representing a significant breakthrough in enterprise adoption.
As enterprises move beyond experimentation, they are realizing that the real differentiator lies in the data that fuels their AI. While most enterprise data infrastructure focuses on structured data stored in SQL databases, GenAI predominantly operates on unstructured data and documents like PDFs and Word files. These sources of data remain largely untapped within enterprises.
In 2024, enterprise data strategies will catch up to meet GenAI's needs. Enterprises will start building data acquisition and processing pipelines for unstructured data, utilizing vector databases and embeddings. Data quality checks to prevent bias will become a priority. Enterprises that can effectively unlock proprietary high-quality data through AI stand to gain a competitive advantage.
Recent advancements have resulted in AI models like GPT-4, trained on massive amounts of data for various tasks. The number of large foundation models (FMs) is rapidly increasing, with the understanding that bigger models gain new capabilities. GPT-4, for instance, boasts 1.76 trillion tunable parameters, and GPT-5 is expected to have approximately 17.5 trillion parameters. However, bigger isn't always better.
Smaller models trained on domain-specific, curated data can deliver comparable performance to their larger counterparts. Microsoft's Phi-2 is a prime example, with only 2.7 billion parameters, yet exhibits performance on par with larger models. Smaller models also require less powerful hardware, bringing GenAI beyond the cloud and into everyday devices such as phones, cars, and medical devices. In 2024, we anticipate an abundance of smaller models that are powerful in specialized tasks.
Commercial models like GPT-4 and Anthropic, deployed on the cloud and accessible through APIs, have gained popularity due to their ease of use, rapid jumpstart capabilities, and pay-as-you-go pricing models. However, concerns surrounding data privacy, security, and intellectual property rights have posed challenges. Additionally, the costs escalate as GenAI applications scale in production with more content and users.
To address these concerns, open-source FMs have emerged. While they may lag in response accuracy and performance, models like Meta's Llama2 have made significant advancements in improving response quality. Open-source models, when self-hosted on an enterprise's own servers, mitigate privacy, security, and IP concerns. Their openness and ability to fine-tune for specific domains make them increasingly sought after. In 2024, enterprises will experiment with small-sized, open-source, self-hosted FMs.
Google's Gemini, a multimodal model, marked a new milestone for GenAI. These models can simultaneously analyze data from various sources – text, code, images, audio, and video. This approach closely resembles how the human brain learns, providing AI with a more comprehensive understanding of context, key for enabling richer responses.
The introduction of multimodal models will drive GenAI from cognitive search to rich content generation. As enterprises across industries explore new use cases made possible by multimodal models, the entire GenAI tech stack will undergo disruption. This shift will entail an upsurge in the development and adoption of multimodal models.
AI went mainstream last year, and technologists unveiled numerous practical GenAI applications. Today, AI use cases permeate every industry, capturing the focus of businesses of all sizes. Looking ahead to 2024, we can expect a surge in new GenAI deployments as enterprises rapidly transition their AI proof of concepts into pilots, utilizing the latest models, patterns, and cloud services.
In conclusion, the widespread adoption of GenAI is on the horizon, and businesses across industries must prepare for this transformative technology. As we move into the next era of AI, staying ahead of the curve and harnessing the power of GenAI will undoubtedly become crucial for maintaining a competitive edge. The future is filled with possibilities, and GenAI is poised to reshape the way we work, innovate, and interact with technology.
