Towards a ChatGPT Maturity Model for Organizations

ChatGPT is marvelous. ChatGPT is beneficial. ChatGPT is concerned. ChatGPT is dangerous. There is a broad spectrum of attitudes, but in the end, it depends on your organization's maturity in working with ChatGPT. We think organizations may find ChatGPT more valuable following the right way of integrating.

Pingping Consultation provides organizations guidance on integrating ChatGPT. The main thing we are looking at is the foundational capabilities of the organizations.

Representation Capability

Representation Capability is the ability to identify patterns in ChatGPT answers and map them to business objects.

ChatGPT answers have a finite number of categories. In our?post on the ChatGPT mechanism, we described that ChatGPT learned to answer from human labelers, and their response is rooted in those human patterns.

Understanding the common ChatGPT answer patterns relevant to your domain is essential for organizations. In addition, the high-frequency patterns are worth analyzing if the answer patterns distribute unevenly like a Zipf law.

Your business may disagree with the ChatGPT answers, but you should learn those ChatGPT answers, distill patterns, and ideally conduct automatic detection. Those ChatGPT answer patterns provide an excellent intent categorization, no matter whether they are right or wrong. And if your organizations want to improve on the bad part, the first step is to recognize them.

In our deep thinking post Integrating ChatGPT with Proof Assistant — State of Affairs, we explored the topic of what's true knowledge, and "being able to uniquely identify patterns" follows Descartes' philosophy on understanding worlds in a reasonable way. This is very practical insight: if organizations want true knowledge, it would be very hard if you do not have a "discrete" element in the representation.

Extension Capability

When your organization has a working technology to recognize relevant ChatGPT answers, the next thing is to build extension capability, which fulfills your organization's will on how the information should be present and how the user-machine interactions will go.

That implies you should have known the existing ChatGPT answer well in the previous "Representation Capability." And that also means you have domain-specific business rules and internal facts that can perform reasoning, identifying the direction for a continuation of an existing ChatGPT answer.

We could reuse traditional Chatbot dialogue for an extension of a ChatGPT answer. However, the best strategy to extend ChatGPT's response is to know the extra information your organization environments hold and to refine the existing ChatGPT answer. Your tasks are diversified. You can do moderation, you can do dialogue customization, or you can connect to your own database.

In our medium post?Marry ChatGPT with a Proof Assistant — A Case Study on Ad Hominem, we showed an example of extension capability: we can generate custom-prompted messages following a pre-defined structure.

Value Attribution Capability

Suppose you have ChatGPT technology to do a similar job to your traditional enterprise knowledge base. In that case, the traditional knowledge base has extensive features to support the value reward. As a result, the content creator gets rewarded, and organizations constantly track content improvements.

The continued operation of knowledge creation, improvements, and deprecation should rely on the existing mature pattern, even though the technology is new. Hence organizations who figured out how that part fits in would have the maturity in this category, the "Value Attribution Capability."

In our medium post?ChatGPT/LLM needs Formal Reasoning for a practical, safe use, we mentioned the reasoning technique should play a role in AI community. It is a similar kind of thinking.?

Conclusion

We proposed a blueprint of the ChatGPT maturity model. Pingping Consultation will later have a more comprehensive publication for a playbook to integrate ChatGPT for organizations.