The effect of ChatGPT on KM

You would have to be living under a rock to have missed the hullaballoo these last few months regarding the release of ChatGPT. According to many, it is the most amazing thing since sliced bread, a sentient, nay superior, intelligence. Although it is a significant technological breakthrough, the hype it is generating in early 2023 is reminiscent of that which surrounded the launch of IBM Watson—a game changer that ultimately failed to change the game, proving costly to IBM both in terms of revenue and reputation.

At this peak of ChatGPT hype, we have to ask what value it may bring. We need to look under the hood to understand the answer to that question. ChatGPT, at its core, is a truly massive, large language model (LLM), one consisting of close to 200 billion parameters.

Large language models, also known as neural network-based language models, have gained much attention recently for their ability to generate humanlike text. These models are trained on massive amounts of text data and use advanced machine learning algorithms to learn the patterns and structures of natural language.

ChatGPT wrote that, and it’s a pretty solid explanation. There is no question that ChatGPT is impressive, writing and understanding language in ways previously unimaginable for a machine. It’s also remarkable in that it utilizes a once unbelievable amount of computing power and, by default, energy, to function. Equally impressive is the fact that it has cost billions of dollars to build, and many more billions of dollars are already earmarked for further investment. Moreover, it is costing around $3 million per day in basic operating costs. If we take into account the fact that it is running on Microsoft Azure pretty much for free, that cost may actually be closer to $8–$10 million per day. These are staggering sums even by Silicon Valley standards.

What this means, in practice, is that the investors will most certainly demand a huge return. There is a desperate hunt on to find sufficient monetizable use cases and customers to make it profitable. That is where problems may arise. If you are an eco-warrior, then ChatGPT is certainly not your friend.

Limitations of ChatGPT

Though impressive, ChatGPT has two significant limitations that it can likely never overcome. The first is that to build such a massive LLM, you need to access vast amounts of data. Oddly enough, that is the simple part; the web is host to vast amounts of data, from news sites and academia to websites and social media. The tricky part is knowing which of the data is accurate. To ChatGPT, it’s just data.

The second problem (as a result of the first problem) is that ChatGPT generates “convincing” outputs, but there is no way of knowing if they are true or not. At its worst, here is one future use of ChatGPT: It is almost ideally suited to deliberately and purposefully generate misinformation. At the core of this problem is an all-too-common assumption in the tech world that more data is good, and less data is bad. As my Deep Analysis colleague Matt Mullen wryly notes, data scientists believe that “Weight=Truth.”

Regardless of the genuine concerns regarding ChatGPT, AI capabilities built from LLMs will play an increasingly prominent and critical role in knowledge work. Hence, anyone involved in the KM world must clearly understand where such tools can add value and where they should be left alone. So to be clear, even though you can have a conversation with or read an impressive essay generated by a language-based AI system, that system, by default, lacks comprehension and ethics and will be riddled with bias. It also lacks common sense. Like an experienced politician, it will always give you an answer to your question, but on later examination, you may find the answer you received is misleading or untrue. So what possible value or role can these systems have in KM and business in general if they are so fundamentally flawed? Well, here’s the kicker: They can offer quite a lot of value, actually.