Generative AI - A basic primer for commercial use

tom benham

tom benham

— 4 min read

What is generative AI?

Generative AI is another name for the large language models (LLM) that power ChatGPT, Claude, Gemini and other LLM solutions. There is a lot of discussion in the industry and in the culture at large about artificial general intelligence or AGI as well, however these are quite distinct.

Generative AI or gen ai, represents a kind of raw intelligence that can be directed at or applied to specific tasks. Of course it can be used like someone uses ChatGPT where you ask questions and receive a response however the focus here is how do we use this in a commercial setting. These tasks do not have to be repetitive or necessarily narrow, however the llm(s) being applied still require a significant amount of guardrails and integration, to extract the value that is on offer. This kind of intelligence is not autonomous. Additionally, generative ai can be applied at increasingly low costs on a per token basis. Token here meaning a single word. For example, openai’s GPT-4o provides its service at $2.50 for 1 million tokens.

AGI on the other hand represents the arrival of something akin to a mature human brain. That is, it can be autonomously directed at a task and then it makes its own decisions about how to proceed in completing that task. It also will have the advantage of being able to learn about any topic much faster than any human could as well as connect to the millions of other AGI, and learn from their learning. It will however, at least initially, be incredibly expensive.

What are the forms of gen ai?

Gen ai is now truly multi-modal. That is, it can interact with, understand and then ultimately generate text, audio, images and video. All of these mediums are relevant to commercial use cases although text is the most predominant at present.

What's the value proposition of generative ai?

If these LLMs are a kind of raw intelligence that can be directed at or applied to specific tasks, the primary value proposition on offer I think is two fold:

  • The complexity of applying scalable machine intelligence has in effect been abstracted away from the underlying modeling and coding. Or in layman's terms, anyone can apply machine intelligence! You don’t need to be a coder.
  • This intelligence can be applied at scale and in coordination with itself and other people.

The practical implications of point 2, is that ai will both replace and augment humans in the work environment. At bottom, a well integrated gen ai solution offers endless possibilities for cheaper “workers” as well as higher quality work in many areas.

The challenge then on the individual or commercial side, in harnessing the value proposition is also two fold:

  • Working out which aspects of a given task or series of tasks or processes, it makes most sense to apply this intelligence to.
  • Applying the intelligence by way of workflow integration including implementing the appropriate guardrails, human touchpoints and the relevant data stores that enhance and direct the applied intelligence.

The ideal and ultimate experience, whether as a customer or internal users, should be one of interaction with an extremely competent coworker, sales rep, advisor or some other externally facing actor. Gen ai is reawakening the old process reengineering skill set, so we can more systematically and seamlessly integrate digital and human intelligence into our workflows.

Another way to think of this is that previously we had very well defined interfaces and distinct boundaries between human and technology, particularly in the work environment. There was a bunch of code and it did stuff for a specific application that we interfaced with deliberately, using very clunky constrained processes e.g. SQL queries. Now however, with the advent of gen ai along with mobile technology and augmented (and maybe virtual) reality, the lines between our physical, mental and digital worlds are blurring. Our interfaces are more natural and interactions almost indistinguishable from that with another person.

What are the different ways then to leverage and integrate gen ai?

Meeting the challenge of the above two questions then is an interplay between experience with the gen ai as well as domain content and process knowledge. Bringing these together is a process.

Putting aside the proceeds and any specific domain expertise, there are different levels of complexity in applying gen ai and as we move down the list the complexity and technical skill required to harness the ai increases. Overall these approaches can be described as:

  • Prompt engineering e.g. defining a role for the ai to play and how to interpret queries
  • Retrieval-Augment Generation (RAG ai). This includes:
    • Deterministic data store integration e.g. SQL databases. In this case it is making specific queries and returning results, that help inform its response,
    • Probabilistic data store integration e.g. vector databases. In this case it is querying text and finding the most similar matches between the query and the text, which then helps inform its response
  • Agentic AI. That is, chaining together the ai in a sequence of steps OR creating an agent environment where there is a controlling ai that has tools at it’s disposal and it coordinates usage of these to generate a response.
  • Model fine tuning e.g. enhance a base model, to be for example a Q&A bot for coding questions.

Each one of these approaches requires a broader discussion and deep dive however some combination of these is usually what is required.

Going forward and enterprise versus build your own adventure

Many start ups are tackling this space and building tools that further abstract away the technical complexity of the above and provide complete solutions for specific workflows e.g. sales, customer support, auditing etc. Whether these plug and play solutions make sense for a company will differ by company.

People should start to make the mental shift towards thinking of gen ai as in essence intelligence and work capacity that can be applied to their company rather than as just another technology implementation. Done well, company users and their customers won’t be able to tell the difference beyond the fact that their experience feels better.

Some good reference articles:

Read more