Are you one of those leaders who fervently discuss the pros and cons of chatGPT in your business? If so, you are on the wrong path, and you are diverting your company and yourself from the important discussions.
A few days ago, I facilitated a discussion around the use of AI in a particular industry. I quickly realized that there is a fundamental problem with the general understanding of what chatGPT is and not, and hence which role it should - and should not - play in any business.
Let's start with clarifying the difference between chatGPT and GPT-4. Let's use a car as analogy. ChatGPT is like the chassis of the car, the interface that you as a driver interact with. Underneath the chassis, there is an engine. The engine in chatGPT is called GPT-4. No matter how cool and fancy the chassis may look, it's the power of the engine that determines the capabilities of your car. In the world of generative AI, it's the same; you interact with chatGPT, but the capabilities of the AI lies in GPT-4, which is the large language model (LLM) underneath the chat interface.
ChatGPT isn't the only "chassis model" out there, though. ChatGPT is the "chassis model" from OpenAI, Bard is the "chassis model" from Google, and Ernie is the "chassis model" from Baidu, to mention only a few. The underlying "engine models" are called GPT-4, PaLM 2, and ERNIE 3.0, respectively.
And this is where your focus should be; on the engines, aka foundation models. Whether you choose proprietary or open source, the foundation models can be accessed through APIs. With a tiny bit of help from your tech resources, your company can build applications that access the foundation model through APIs. Then you tune the model on your internal data and context, and put up guardrails that are relevant for your specific situation.
I'll use another analogy. Using chatGPT( and peers) is like working at the public library. The Deichman library in Bjørvika in Oslo is a beautiful and inspiring place, and I love roaming the bookshelves and drawing in the vibrant atmosphere. But would you put your company secrets on the shelves of the public library, and make confidential calls in the open working space? Would you contemplate making the public library a part of your expansion plans? Of course not! For some purposes working in the inspiring environment of Deichman can indeed be highly productive and very useful. However, it is not a relevant element in your corporate strategy.
Next time someone in your company diverts the strategic discussion around AI with pros and cons of chatGPT, invite them for a coffee in the nearest public library.
ChatGPT may very well be a useful tool for yourself and for your employees. Just like a Remarkable is a great tool for taking notes, Excel is a great tool for building business cases, and Bose headphones are great tools for reducing noise. These tools, however, have no strategic value. The strategic value lies in the utilization of Large Language Models as the foundation for your context-specific applications. From now on, discuss the engine, not the chassis.
Comments