A Starry Night for AI

AI as a Force Multiplier for the Business

In swirls of night, Van Gogh did paint,
A starry sky, a world untaint.
The AI too, in code and line,
Can weave its dreams, in data mine.

But what’s an art, without a soul?
A canvas blank, a scattered goal.
In business too, the AI thrives,
When purpose guides, and dreams it drives.

Author: ChatGPT (v4)

We noted in our last post that McKinsey projects a massive productivity boost from generative AI that will unfold over the next 20 years. With apologies to Van Gogh, we concluded that, like a Starry Night, the sweeping swirls of the report can fire the imagination, but much finer brushstrokes will be needed by a company to gain tangible benefits. We recommended in our post to begin with narrowly defined prototypes to understand the benefits and risks of generative AI for your business. In this post, we want to explore that a little further.

Microsoft Research published a report which expanded on their technique, called Grounding, for communicating with an LLM model. The idea is to anchor an LLM in a specific use case by feeding it information that is relevant to the process you are looking to address. While LLMs come with a vast amount of knowledge already, this knowledge is limited and not tailored to specific use-cases. To obtain accurate and relevant output, we must provide LLMs with the necessary information. In other words, we need to "ground" the models in the context of our specific use-case.

LLMs are not databases, even if they possess a wealth of knowledge. They are designed to be used as general reasoning and text engines. We have been saying that from the beginning: while content creators gain readily from unfiltered interactions with LLMs, the most productive use cases for a business resides in customer, supply chain, financial and other operational systems of the company. LLMs need to be harnessed to operate as logic models, integrated with core business processes. As companies reimagine their business processes with AI integrations, we expect a few ‘sweeping swirls’ will be made manifest, opening new opportunities to ‘obliterate’ the old, and deploy the exceptional. Microsoft research describes a pattern called Retrieval Augmented Generation (RAG), which means pulling in the relevant information to the LLM that is specific to the context. While we subscribe to that pattern for certain use cases, we think a more powerful force multiplier is to pull in the relevant function which retrieves the requisite information.

Imagine for a moment the Starry Night we envision: A business has hundreds if not thousands of APIs which access extraordinary sets of functions for running the business. The functions access real-time databases, with current information on product inventory, pricing, shipping, warranties and other data, driving sales and revenue. Rather than feeding information to the LLM, feed it information about the APIs, and let the LLM manage the workflow.

If you are wondering what this looks like in code, here is an example:

type FunctionNames = 'get_products' | 'get_pricing' | 'get_inventory_availability' | 'checkout'

export const functions: {
  name: FunctionNames,
  description: string,
  parameters: object,
}[] = [
  {
    name: 'get_product',
    description: 'Get the set of products which most closely match the user query.',
    parameters: {
      type: 'object',
      properties: {
        limit: {
          type: 'number',
          description: 'The number of products to return. Defaults to 25.',
        },
      },
      required: [],
    },
  },
  // other functions follow here
]

We like the direction that OpenAI is taking with GPT4, which has the superpower to call functions and incorporate the results with user interactions. We described another superpower in a prior post called plugins, permitting companies to incorporate LLMs into their production systems. With these advancements in LLMs, we not only envision Starry Nights, but practical grounding for business cases that deliver 'needle moving' productivity for the company.

Give us a call. We’d like to discuss how we can help you compose an AI masterpiece for your business.