Fireflies After the Storm

Agentic Architecture

We thought the firefly was a fitting metaphor for the new class of LLM models hitting the market. Early AI days were dominated by giants like OpenAI’s GPT-4—costing over $100 million to develop and packing a trillion parameters—a storm of complexity for any executive navigating fast-paced markets. But the next wave of AI innovation is all about thinking small. This wave might not be as serene as a firefly on a summer evening, but you get the idea. Welcome to the era of specialized, small-to-medium language models. These leaner, more efficient models are trained on narrower datasets and designed for specific tasks, offering a cost-effective alternative to the behemoths that have dominated headlines. In a world where GPT-4-like models require vast computing resources and high operational costs, smaller models are proving to be a game-changer, especially for businesses looking to harness AI without breaking the bank.

Why Smaller is Smarter

Tech giants like Microsoft and Google, as well as a host of innovative startups, are betting on these smaller models for good reason. Microsoft’s small model, Phi, is reportedly 1/100th the size of the free model behind ChatGPT, yet it performs many tasks nearly as well. The key? Specialization. These models are fine-tuned to excel in specific domains—be it legal, financial, or customer service—allowing them to deliver results at a fraction of the cost. For instance, a small model trained on internal communications or sales data can deliver the same level of accuracy and relevance as a large model, but at one-sixth the cost per query. This shift not only makes AI more accessible but also more practical for businesses that don’t need the full capabilities of a model like GPT-4 for everyday tasks.

The Strategic Advantage of Specialized AI

As the market for AI matures, the focus is shifting from grand, science-fiction-like demonstrations to practical, business-oriented applications. Companies are increasingly looking to develop AI models that are not just powerful, but also efficient, cost-effective, and tailored to their unique needs. We like to think of this as building and deploying (recruiting?) a roster of agents for your production network, highly specialized apps and functions centered on tiny language models for specific domains or topics.

Take Adobe’s Firefly, for example—a generative AI model designed specifically for commercial use, trained on a library of licensed stock images. Or consider Intuit’s model for financial management, which benefits from years of interactions with business customers. These models are not only more secure and reliable, but they also offer businesses a competitive edge by being tailored to their specific industry and use cases.

Preparing for the Shift to Specialized AI: 5 Key Steps

For business executives looking to stay ahead in this evolving AI landscape, here are five steps to prepare for the rise of specialized AI models:

1. Leverage Your Data: Identify key datasets within your business that can be used to train specialized AI models. Your internal data—ranging from customer interactions to operational metrics—can be a goldmine for creating models that deliver tailored solutions. But be aware that curating and remediating your inhouse data will take time and resource, but the payoff is worth the investment.
2. Identify Customer Touch Points: Pinpoint areas in your customer journey where specialized AI can be deployed to enhance experiences, improve service, or streamline processes. Whether it’s customer service, marketing, or financial advice, specialized ai agents can offer significant value.
3. Upskill Your Workforce: Equip your teams with the skills needed to build, fine-tune, and deploy AI models. Investing in AI literacy and technical training will ensure your organization can fully capitalize on the potential of these models.
4. Optimize for Efficiency: Build the business case, identifying where value will be gained from model adoption for your customers and your operations. The deployment of GenAI models in your business and production systems is a new way of operating, and requires forward thinking to realize the value.
5. Partner with AI Innovators: Collaborate with AI vendors and startups who specialize in building and deploying smaller models. These partnerships can provide you with access to cutting-edge technology without the need for massive investment in AI infrastructure. Partnerships with firms like Hugging Face, NVIDIA, Amazon Bedrock and others are worth exploring to provide resources to develop your specialized models.

As the AI landscape shifts, the move towards smaller, specialized models is more than just a technological evolution—it's a strategic leap forward. By preparing now, your organization can harness the efficiency, cost-effectiveness, and precision of the next generation of AI. When the storms clear, the fireflies won't just flicker—they'll swarm, lighting up the market with a new wave of innovative services and products powered by these specialized models. Get ready for the small model revolution.