The Rise of Logic Models

A Reasoning Engine for Business Applications

It is no surprise that Large Language Models have yet to impact business in any meaningful way. Sure, the frenzy over LLMs is palpable but the integration into production systems of most companies is negligible. Our perspective, published in earlier posts, is that most of the impact of LLMs has been isolated to repetitive, low-stake tasks like copywriting, blog posts and email generation. LLMs are creative machines, not precision machines. If you ask an LLM for a price on a product from your catalogue, which we have tested at scale, you may be surprised by the answer. And ‘surprise’ is never an acceptable answer in production systems.

Andreessen Horowitz recently published their point of view on the adoption cycle for LLMs. They focused on applications in the B2B world and believe that while ‘Wave 1’ has been about content creation, ‘Wave 2’ will have a much more meaningful impact through content synthesis. In other words, applications will be deployed by companies which ingest large volumes of information and create cogent summaries for decision makers. We agree that there is tangible value in that use case, but, with apologies to a16z, we think the real Wave starts with 3.

We’re early in the process but envision this design point. Rather than generating content or summarizing documents, the LLM executes a sequence of tasks correctly by reasoning through the shared relationships among resources, activities, outcomes and instructions. The Language Model becomes a Logic Model. Integrations can be discovered and consumed by the models in response to events, such as new orders, warranty claims, product search or service requests. Rather than indexing and exposing the proprietary data of the firm, models are trained on the meta-data of applications, anchoring the ‘reasoning engines’ with instructions about how to access and use apps. These instructions are not unlike those found on a public web site, helping users to navigate the process for buying and shipping products, for example. You get the picture. Modular. Secure. Reusable. Maintainable. Intelligent. And we think affordable.

We are mid-stream testing the concept and like what we see. With reasonable precision, and ‘human-in-the-loop considerations’, we anticipate a path for technology adoption that could bring remarkable transformation to a company’s operations. Application delivery will transform as well.

At Strategic Machines, we hope we can give you a few reasons to engage our team on your next AI project. Let’s connect!