OpinionA future beyond models
Opinion
A future beyond models
"The AI market is evolving rapidly and encountering obstacles along the way—such is the nature of a technology this transformative," writes Ori Goshen, co-founder and co-CEO of AI21 Labs.
Organizations understand that implementing AI effectively tcan lead to significant improvements in workflows, resource efficiency, and productivity. Beyond this, the market recognizes that AI not only enhances existing performance, but also creates new opportunities to innovate and strengthen competitive advantages. Over the past two years, businesses have come to a stark realization: : those who don’t adopt AI today will find themselves at a disadvantage tomorrow, likely facing significant competitive challenges.
The Limits of Language Models
After the initial excitement, it has become clear that language models alone cannot fully serve organizational needs. They often lack the accuracy and reliability required for handling professional or sensitive information. These models struggle with grasping broader context, are limited in their knowledge, and are inherently unstable - since they are fundamentally statistical models. Additionally, they may not always stay updated with real-time organizational information and frequently encounter challenges with complex tasks.
It’s important to note that expectations for these models are sometimes unrealistic. We’re still in the early stages of the AI revolution.. Even the most advanced technology may not produce flawless long-form content or perfect code on the first attempt without external inputs or multiple drafts.
The Rise of Compound AI Systems
The good news is that we are entering the era of Compound AI Systems—sophisticated systems that utilize several components to solve tasks, rather than relying on a single action from one language model. These systems utilize multiple language models, each specializing in a different aspect of the task. They also integrate external tools—such as search engines, databases, or specialized functions that can be activated as needed—alongside system logic components that oversee execution sequences, perform validation, or determine the optimal strategy for completing tasks.
A Paradigm Shift
The progress we see in the AI market isn’t driven solely by groundbreaking technological advances or significant improvements in model quality. Rather, it’s a conceptual shift: abandoning the notion that a single query to a language model can solve all problems. Recognizing that language models alone are insufficient has paved the way for the development of multi-component systems that integrate models, tools, and processes to form the foundation of AI’s future.
One such approach is RAG (Retrieval-Augmented Generation). RAG retrieves relevant information from databases and produces more grounded answers. These systems don’t merely rely on the internet; they use focused data sources to make the responses more reliable.
The Effectiveness of Complex Systems
Compound systems aren’t necessarily "smarter" than standalone language models, but they are significantly more effective due to their ability to combine multiple components and tackle multi-step problems more reliably, offering organizations greater control over outcomes.
The Challenges of Complexity
However, compound systems are complex and come with their own set of challenges. Developing and implementing compound systems requires seamless coordination among components, ensuring quality and performance while minimizing disruptions to the overall system’s functionality.Unique obstacles for AI systems include data privacy concerns, model stability, and adjustments to required organizational changes. Additionally, transitioning to these advanced systems demands a shift in mindset, a thoughtful transition to automation and proper oversight of model use. Compounding these challenges is a shortage of skilled personnel capable of designing and maintaining such systems effectively.
The Future of AI Systems
Despite these hurdles, compound systems provide stability, control, and reliability—critical elements for large organizations. AI is becoming essential for competitive survival, and these systems are poised to solve organizational challenges while driving improved business outcomes.
In the near future, discussions about language models in isolation will likely fade. Instead, the focus will shift to compound systems, which will become increasingly sophisticated. Only a handful of companies may fail to adopt these advanced systems.
The AI market is evolving rapidly and encountering obstacles along the way—such is the nature of a technology this transformative. But the trend is unmistakable: the world—and the business sector—will not stop advancing.
Ori Goshen is the co-founder and co-CEO of AI21 Labs
