Balancing Specialized Task-Based AI with the Flexibility of LLMs in Modern Enterprises
Remember a time, not so long ago, when AI in enterprises primarily focused on building models for specific tasks like loan approvals or fraud protection? This approach, often termed "good old-fashioned AI," has been the backbone of AI applications in various industries. However, the advent of Large Language Models (LLMs) like ChatGPT has shifted the focus towards more generalized AI models.
Despite this shift, task-based models haven't become obsolete. They continue to be a critical component in the enterprise realm, effectively addressing a wide array of business challenges. Werner Vogels, Amazon's CTO, emphasizes the ongoing relevance of these models, particularly in solving real-world problems.
LLMs, like Google's PaLM 2 and Meta's LLaMA 2, have indeed revolutionized the way we interact with technology. Their ability to understand, generate, and interact using human language has opened new frontiers in AI applications. But this doesn't diminish the importance of task-specific AI models, which are designed to handle specific tasks within their defined boundaries, unlike LLMs that possess a broader scope of application.
"Task models... can be more performant because they’re designed for a specific task" - Jon Turow, Partner at Madrona.
One significant aspect of LLMs is their cost. They require extensive pre-training, often costing millions of dollars, and utilize vast repositories of textual data to gain foundational knowledge. This is where task-based models gain an edge – they are often smaller, faster, cheaper, and in some cases, more performant due to their specialized nature.
"Good old-fashioned AI... is still solving a lot of real-world problems" - Werner Vogels, Amazon CTO.
Moreover, the rise of LLMs doesn't eclipse the role of data scientists. Previously, when task models dominated the AI landscape, teams of data scientists were essential in developing these models. Their role remains crucial in the era of LLMs, as they continue to evaluate and understand the intricate relationship between AI and data within organizations.
Both task-based AI and LLMs are expected to coexist in the foreseeable future. LLMs offer unmatched flexibility and applicability, but task models retain relevance due to their specialization, cost efficiency, and performance advantages.
